Introduction

Wireless communications have made remarkable strides with the advent of 5th generation (5G) technology, enabling unprecedented levels of connectivity and opening up new possibilities for innovation and progress in various fields. Following Edholms’ law, data traffic has doubled every 18 months over the last three decades. This trend predicts a data speed requirement of terabit per second (Tbits−1) before 20351 to meet the rising demands. This has led to the development of 6G, which targets extremely high data speeds ( ~ 1 Tbits−1), ultra-low latency (below a ms)2, and advanced wireless connectivity, which are mandatory in the fields of medical robotics, surveillance drones, and digital-physical twins3,4. The transition from 5G to 6G will include the increase of carrier frequency from microwave to certain bands in the terahertz (THz) range, as this offers higher bandwidth and faster data transfer rates5,6,7,8,9,10. Operation at sub-THz frequencies (0.2–0.3 THz) is considered optimal because of the significant atmospheric attenuation above 1 THz.11,12.

Receivers play a crucial role in wireless communication approaches for detecting data signals. State-of-the-art sub-THz receivers can loosely be categorized into electrical and optical receivers. Electrical based schemes such as heterodyne detection13,14,15 achieve data rates in the 120 Gbits−1 range5,14,16,17,18. They are typically constrained by their relatively non-flat frequency response, bandwidth and non-linearities due to the employed mixers. On the other hand, the optical receivers employ Schottky barrier diodes (SBD) to detect the sub-THz radiation and directly convert it to the baseband. Such devices have been used to demonstrate  ~ 100 Gbits−1 data transmission19,20,21,22,23. However, their conventional maximum response rate is limited to 40 GHz24, and they rely on off-chip focusing elements such as hemispherical silicon lenses21,24, horn antennas with waveguide housings19,20,23, or THz amplifiers19,25 to efficiently couple free-space THz radiation, thereby increasing the overall footprint. Additionally, SBDs typically use air-bridge structures, which add fabrication complexity and limit compatibility with monolithic or CMOS-integrated platforms. Recently, photonic receivers based on high-speed plasmonic modulators have been demonstrated. Here the modulator offers a bandwidth up to 500 GHz, and beyond26,27,28,29,30,31; however, it still depends on the rest of the receiver architecture. For example, the plasmonic modulator encodes the data from the sub-THz carrier onto a telecommunication optical carrier to detect the data signal with a conventional photodetector. High data rates  > 100 Gbits−1 over distances  > 1 km have been demonstrated with this approach. See Table I for a summary of key parameter values.

While all the above-mentioned approaches can operate in the sub-THz regime and achieve high data rates, they still involve significant complexity and require external components to complete the receiver scheme (such as harmonic mixers, local oscillators, amplifiers, additional photodetectors, etc.) to facilitate the conversion of sub-THz radiation into electrical signals. This leads to challenging integration and widespread deployment of these receivers in terms of footprint and power consumption. These limitations prevent the scalability and diverse integration of these receivers. Moreover, none of the current techniques simultaneously offer the following desired characteristics32,33: low noise, 50 Ω output impedance, compatibility with complementary metal-oxide-semiconductor (CMOS) technology34, low cost, small footprint, and operation without the need for additional external components to complete the receiver functionality.

Here, we demonstrate that graphene-based receivers are promising candidates for 6G wireless technology, as they simultaneously meet all the above-mentioned requirements. In particular, we achieve the detection of data streams using sub-THz graphene receivers. To accomplish this, we integrate multiple concepts into a single device, including the use of a resonant sub-THz on-chip antenna, a high quality-factor cavity with a back mirror, a high-mobility graphene channel that also exhibits reduced resistance, closely matching the 50 Ω impedance of the measurement electronics, and a fast electrical readout comprising a microstrip transmission line. The incorporation of these elements enables both high responsivity and fast response times, which are crucial for real-time wireless data detection in the sub-THz band.

The capabilities of this sub-THz graphene direct receiver are highly promising. Our simulations predict that a 3 dB bandwidth exceeding 300 GHz, responsivities of approximately 1 A/W (120 V/W) and a noise-equivalent power (NEP) of 14 pW/\(\sqrt{{{{\rm{Hz}}}}}\) are attainable. These predictions are supported by experimental results, where we achieved a multigigabit per second data rate over a distance of approximately 3 m between the transmitter and the graphene receiver. We demonstrated a crucial trade-off between bandwidth and responsivity, as observed from measurements across different devices: a setup-limited 3 dB bandwidth of 40 GHz was achieved in a low-responsivity device, while a maximum responsivity of 0.16 A/W (30 V/W) and a noise-equivalent power (NEP) of 58 pW/\(\sqrt{{{{\rm{Hz}}}}}\) were measured in devices with a lower bandwidth of approximately 2 GHz. These results highlight the potential of the receiver for applications such as chip-to-chip communication and close proximity device-to-device communication.

Results

Receiver operation principle

Figure 1a shows the graphene-based receiver, which consists of an hBN-encapsulated graphene. The optimized geometry comprises a short and wide graphene channel (see Methods and Supplementary Notes 1–4 for fabrication and simulations details) that simultaneously exhibits an enhanced photoresponse and a reduced resistance that closely matches the 50 Ohm impedance of the measurement electronics. This impedance matching is critical for efficient data stream detection, as it minimizes signal reflections and ensures effective power transfer. The receivers operate at zero-bias by exploiting the photothermoelectric (PTE) effect35,36,37,38,39, where the sub-THz radiation absorbed by graphene generates a photoresponse driven by a temperature gradient, provided there is doping asymmetry in the channel37,40,41,42,43,44,45. In our implementation, the doping asymmetry across the graphene channel is induced using a split-gate structure. Independent voltages are applied to the left and right gates to create a graphene pn-junction, thereby maximizing the Seebeck coefficient difference between the n-doped and p-doped regions and, consequently, enhancing the photoresponse37,41, as detailed in Supplementary Notes 4 and 5.

Fig. 1: Sub-THz digital data-stream detection with graphene-based receiver.
Fig. 1: Sub-THz digital data-stream detection with graphene-based receiver.
Full size image

a Schematic representation of a typical configuration of receivers with incident sub-THz radiation including the device circuitry. The illustration contains the back mirror, Si/SiO2 substrate, graphene encapsulated with dielectrics (e.g., hBN, alumina), and antenna/split-gate. b Experimental setup used to detect the data stream. In this case, we employ a near-field probe to compensate the sub-THz beam divergence. The abbreviations for the different elements of the setup are described in the main text. The panels in dashed lines show the corresponding frequency of the utilized signal in each component, where purple, orange, and green correspond to the electrical (radiofrequency, RF), sub-THz, and telecom frequencies, respectively. c Eye diagrams of NRZ OOK data streams at 1 Gbits−1 at different output powers of the transmitter (estimated at the output of the multiport power amplifier, MPA) of 14.5 dBm (top panel) and 10.5 dBm (bottom panel). The carrier frequency is 0.213 THz. d The signal-to-noise ratio (SNR) and bit error rate (BER) extracted from the collected eye diagrams at 1 Gbits−1 at different output power of the transmitter.

This split-gate simultaneously serves as a half-wavelength dipole antenna, with a length of 600 μm, to attain a resonant frequency of  ~ 0.230 THz. The sub-THz radiation is highly concentrated at the  ~ 400 nm gap of the antenna, which overlaps with the photoactive area (i.e., the graphene pn-junction)37. In conjunction with this, we employ a back mirror below the highly resistive silicon substrate, which is transparent in the sub-THz range and has a thickness of  ~ 300 μm, as shown in Fig. 1a. By following this approach, we achieve multiple reflections between the back mirror and antenna, where the graphene monolayer is located to significantly boost the absorption and photoresponse, ensuring an efficient data stream detection. Thus, the integrated antenna and a back mirror overcome the low graphene absorption and large mismatch between the small optically active area ( ~ μm2) with respect to the comparatively large spot size at sub-THz wavelengths ( ~ mm2)37,46,47.

These receivers can operate in a passive mode by exploiting the PTE effect in graphene at room temperature and under zero-bias operation35,37. This approach ensures zero dark current, direct photovoltage or photocurrent readouts, low flicker noise, and compatibility with CMOS platforms, as graphene can be integrated at the back end of the line34,48. The direct detection scheme enables incident data with a sub-THz carrier frequency to be directly converted into electronic signals, without the need for a local oscillator or an upconversion process. Consequently, the above-mentioned characteristics allow for simple circuit design and low power consumption, marking a significant milestone in the field of wireless communications.

We have fabricated 7 devices, where 2 are based on exfoliated graphene and 5 on graphene grown by chemical vapor deposition (CVD); see more details in Supplementary Notes 1–2. We also prepared an evaluation kit with a packaged THz receiver containing one of the devices based on exfoliated graphene. We will first discuss the results obtained with the device based on exfoliated graphene labeled as Device 1 (D1).

Wireless data stream detection

We evaluate the data stream detection of Device 1, which consists of hBN encapsulated exfoliated graphene, as depicted in Fig. 1a. Figure 1b shows the experimental configuration that we used for sub-THz data stream detection (more details in Supplementary Note 6). We use two lasers, denoted as lasers 1 and 2 in Fig. 1b, which operate at distinct frequencies (f1 and f2, respectively) in the near-infrared range ( ~ 1550 nm) and are combined using a 3 dB splitter. We control the detuning of these two lasers (f1 − f2  ≈ 0.213 THz) to generate a beating tone, which in the high-speed uni-traveling carrier photodiode (UTC-PD) is detected29 and is transmitted to a near-field probe. The fabricated sub-THz receiver is placed at a distance of a few centimeters from the near-field probe output. We employ a near-field probe to get more radiation onto the device, because the sub-THz beam is strongly divergent. The output signal from the graphene receiver was amplified by a voltage amplifier with a variable gain ranging from 30 to 70 dB and a 3 dB bandwidth of 1.2 GHz and a real-time digital sampling oscilloscope (DSO) was employed to record the sampled waveforms.

To encode data onto the sub-THz carrier, we use a modulator and an arbitrary waveform generator on the optical carrier generated by laser 2, see Fig. 1b. We encode a random bit sequence with non-return-to-zero on-off keying (NRZ OOK) modulation format at 1, 1.5, and 2 Gbits−1. The collected eye diagrams are shown in Fig. 1c and Supplementary Note 7. The completely open eye diagram at 1 Gbits−1 shows that we successfully convert sub-THz wireless data-streams into electronic signals. The eye becomes narrower upon increasing the bit rate above 1 Gbits−1, and it completely closes for a bit rate of 3 Gbits−1, as shown in Supplementary Note 7. This is setup-limited and arises from the 1.2 GHz bandwidth of the voltage amplifier connected to the DSO.

Subsequently, we measure the eye diagrams for different input powers (measured at the output of the multiport power amplifier, MPA), which is achieved by tuning the variable optical attenuator (VOA) placed after the Erbium-Doped Fiber Amplifier, as shown in Fig. 1b. We observe that the amplitude of the eye decreases with power, as shown in Fig. 1c. The extracted bit error rate (BER) from the detected data streams at different powers is shown in Fig. 1d. The BER reaches a minimum value of  ~ 2.5 × 10−6. This minimum BER value was setup limited by the memory of used instruments. For the signal-to-noise ratio (SNR) as a function of power, we achieve a maximum value of  ~ 13 dB and a minimum of 4 dB for the lowest power. These results were obtained while maintaining a constant voltage amplifier gain of 40 dB. In Supplementary Note 7, we also evaluate the BER and SNR as a function of the amplifier gain by varying it from 30 to 70 dB. Knowing these parameters, such as the optimum amplification gain, enables the design of evaluation kits or future prototypes with an integrated amplifier featuring optimal gain for our sub-THz receivers.

Wireless link performance with evaluation kit

Following the assessment of the initial data stream detection measurements at sub-THz carrier frequencies, we developed a graphene-based wireless receiver evaluation kit for the sub-THz range and used it to demonstrate a wireless link over longer distances. The evaluation kit receiver consists of a printed circuit board (PCB) with an integrated hBN-encapsulated exfoliated graphene receiver, similar to that of Device 1. In this configuration, the receiver is positioned at the center of the PCB (as indicated in Fig. 2a and Supplementary Note 8), with DC pin heads for the split gates, and the ground plane is situated on the right side of the PCB. From the middle to the bottom of the PCB, the microstrip transmission line, designed to have a characteristic impedance of 50 Ω, carries the photogenerated signal that terminates with a Sub-Miniature version A (SMA) connector. A metal block serves as both a support and back mirror (see Fig. 2a).

Fig. 2: Evaluation kit and digital data-stream detection as a function of the sub-THz wireless link distance.
Fig. 2: Evaluation kit and digital data-stream detection as a function of the sub-THz wireless link distance.
Full size image

a Schematic of the configuration containing polymethylpentene (TPX) lenses with a variable distance between the transmitter and the graphene receiver. On the bottom part, the optical image shows a side view of the fabricated PCB-integrated receiver mounted on a metal support, including DC pins and a SubMiniature version A (SMA) connector. The inset shows the optical image of the graphene receiver. b Collected eye diagrams at data rates of 0.4 and c 3 Gbits−1 with a distance of 0.5 m between the lenses and 0.64 m between the transmitter and receiver. The carrier frequency is 0.209 THz. d Bit error rate (BER) in log scale as a function of data rate detection for different distances of the sub-THz wireless link.

Using a similar NRZ OOK amplitude modulation scheme, with a carrier frequency of 209 GHz, we study data transfer using the evaluation kit. We implement three main differences compared to the previous measurements: (i) the use of two polymethylpentene (TPX) off-chip lenses to collimate and focus the sub-THz radiation emitted from the horn antenna with a gain of 31 dB, replacing the near-field probe to enable measurements at larger distances between the transmitter and the graphene receiver; (ii) with the PCB configuration, we avoid the use of RF probes and potential light reflections produced by them; and (iii) we vary the distance (d) between the two TPX lenses from 0.5 to 2.5 m as shown in Fig. 2a. The corresponding free space losses are 42, 50 and 54 dB for 0.64, 1.64, and 2.64 m respectively, which correspond to the distances between the transmitter and the graphene receiver with a dipole antenna gain of 2 dB. This setup allows the examination of data transfer as a function of distance and compensates for the large divergence of the source, as shown in Supplementary Note 9. We emphasize that the use of off-chip lenses is a standard procedure in the field for determining wireless communication link performance at longer distances19,23.

In Figs. 2b, c, we show the collected eye diagrams for the data bit rates at 0.4 and 3 Gbits−1 for a wireless link with a distance of 0.5 m between the lenses, corresponding to a distance of 0.64 m between transmitter and receiver. Fig. 2d summarizes the BER measured with respect to data rates, clearly exhibiting an increase of BER and a drop in the maximum measurable data rate upon increasing the distance (d) between the transmitter and graphene receiver. These results indicate a reduction in the sub-THz incident power owing to free-space losses and beam divergence, as shown in Supplementary Note 9. Therefore, a higher BER is observed even for small bit rates at the largest measured distance. Consequently, it is more difficult to measure higher bit rates at larger distances. A potential solution to overcome this limitation is to further enhance the responsivity of graphene receivers, implement arrays of graphene receivers, and integrate more optimized sub-THz cavities. In the following sections, we explain how to increase the responsivity of devices and if this leads to bandwidth limitations. We note that we measured a maximum data rate of 3 Gbits−1 at d = 0.5 m, and up to 0.5 Gbits−1 at a distance of 2.5 m. In the case of BER, we achieved minimum values of  ~ 2 × 10−5 and ≤3 × 10−2 at a distance of 0.5 m and 2.5 m respectively.

Scalability and responsivity of the sub-THz receivers

Now that we have demonstrated a sub-THz wireless link using a graphene-based receiver evaluation kit, we discuss how these devices can be improved and made scalable. We have fabricated 4 additional CVD graphene-based devices by following the same design as Device 1. We use various dielectrics or bottom encapsulants such as alumina, hBN, or transition metal dichalcogenide (TMD) to evaluate their photodetection performance. All the characteristics of the devices are described in the Supplementary Note 2 and Methods.

Figure 3a shows a summary of the measured responsivity (in A/W units) of the fabricated sub-THz receivers as a function of the graphene mobility. We note that the higher mobility of graphene enables higher responsivity and lower noise-equivalent power (NEP) because of the improved electronic properties and lower residual doping (n*), which leads to a larger Seebeck coefficient38,41,49. As a result, we enhance the PTE effect and overall photoresponse in graphene37,38.

Fig. 3: Spectral responsivity of the sub-THz receivers.
Fig. 3: Spectral responsivity of the sub-THz receivers.
Full size image

a Responsivity of the investigated devices as a function of the charge carrier mobility of graphene. The highest responsivity is observed for the hBN-encapsulated exfoliated graphene receiver, which exhibits the highest mobility. The data points are clustered for receivers based on CVD graphene (green circle) and hBN-encapsulated exfoliated graphene (light blue). b Optical simulations of the field intensity at the graphene channel with and without the back mirror (bm) within the measured spectral range. c Responsivity spectra of Device 2 (D2), which is based on CVD graphene with alumina as the dielectric. The highest responsivity is achieved in the presence of the back mirror and with the incident sub-THz radiation polarization (pol) parallel to the antenna main axis.

Consequently, the hBN-encapsulated exfoliated graphene receiver exhibits the highest external responsivity of 0.16 A/W (30 V/W) with a low NEP of 58 pW/\(\sqrt{{{{\rm{Hz}}}}}\), limited only by Johnson noise due to the zero-bias operation37,43. These results are consistent with the calculated responsivity values (see Supplementary Notes 3–4). The Johnson-noise-limited NEP is described in the Methods section. It is important to note that the experimental noise from the electronics in our unoptimized setup is two orders of magnitude higher than the expected Johnson noise, which leads to an experimental NEP of 1 nW/\(\sqrt{{{{\rm{Hz}}}}}\), as detailed in Supplementary Note 9 and the Methods section. On the other hand, the receiver containing scalable materials (e.g., Device 2, D2) shows a responsivity of 13 mA/W (7.1 V/W) with a Johnson-noise-limited (experimental) NEP of 416 pW/\(\sqrt{{{{\rm{Hz}}}}}\) (12.5 nW/\(\sqrt{{{{\rm{Hz}}}}}\)). We stress that the detection of data streams by CVD graphene-based receivers is not yet achievable owing to their lower responsivity. These results demonstrate that the high quality of graphene significantly enhances its photoresponse in the sub-THz range37.

Spectral photoresponse of the receivers

We now discuss how we ensure an enhanced absorption of the incident sub-THz radiation. We simulate our entire device using finite-difference time-domain (FDTD), as described in Methods and Supplementary Note 3. The resonant peak is spectrally located at approximately  ~ 235 GHz, corresponding to a wavelength of  ~ 1.2 mm. This resonance is expected based on the half-wavelength dipole antenna with a length of 600 μm. From simulations with and without the back mirror, we find that the field intensity increases by a factor of  ~ 6 when the back mirror is integrated into the device, as shown in Fig. 3b.

To evaluate the enhancement provided by the sub-THz cavity, we then measure the frequency-dependent photoresponse of a CVD graphene-based receiver with alumina as the dielectric (labeled Device 2, D2), as depicted in Fig. 3c. We notice that when having the incident radiation perpendicularly polarized to the main axis of the antenna, we obtain a very low responsivity below 1 mA/W at  ~ 220 GHz. However, when exciting the dipole antenna with the incident polarization parallel to its main axis, we achieve an increase by a factor of 15. The responsivity is further enhanced by placing a metallic back mirror below the Si substrate, thus attaining a boost by factors of 4 and 60 compared to the cases without a back mirror and when the antenna is either excited or not, respectively. The spectral position of the resonant peak and the enhancement provided by the back mirror are in very good agreement with the simulation results. The slight spectral shift of  ~ 10 GHz in the experimental resonance peak position relative to the theoretical prediction is attributed to a minor deviation in the silicon substrate thickness, on the order of 10 μm, as detailed in Supplementary Fig. 8. The NEP spectrum is shown in Supplementary Note 9.

Bandwidth measurements at different carrier frequencies

Finally, we study the intrinsic speed limitations that affect these graphene-based sub-THz receivers, including the setup electronics, device resistor-capacitor (RC) time constant, and sub-THz antenna-cavity response. By considering the resistance (R) and capacitance (C) of the devices and the equivalent circuit39,50, we can determine the expected RC-limited 3 dB bandwidth using the following equation:  ≈ 1/(2πRC)39,41,50, where the obtained values are  > 50 GHz (see Supplementary Notes 10–11).

On the other hand, the optical limitation of the 3 dB bandwidth can be calculated by the equation \(\approx \frac{c}{n2LQ}\), where c is the speed of light, n is the refractive index of the substrate (3.42 for silicon), L is the length of the antenna, and Q is the quality factor derived from the resonance width51,52. Considering an antenna length of 600 μm and a Q-factor of approximately 15 for the spectral photoresponse peak (see Fig. 3b and Supplementary Notes 3 and 11), which corresponds to a bandwidth of  ~ 5 GHz. We note that a similar optical bandwidth limitation is observed for a graphene receiver coupled to a microring resonator at telecom frequencies53. We describe how to overcome this limitation in Supplementary Note 12.

To evaluate the 3 dB bandwidth of the fabricated receivers as a function of carrier frequency, we use a setup similar to that described in Fig. 1b (see Supplementary Note 6 for further details of the setup). In Fig. 4a, the sub-THz receivers show a limited 3 dB bandwidth ranging between  ~ 1 and 2 GHz at frequencies between 0.21 and 0.25 THz. In addition, by varying the gate voltage (carrier density), which affects the graphene’s resistance and capacitance,50 we do not observe significant changes in the bandwidth values, as shown in Supplementary Fig. 29. However, when testing the receiver integrated with the PCB (Device 6) at a wavelength of 1550 nm (193.4 THz), it shows a bandwidth of 10 GHz, which is limited by the microstrip line of the PCB, as shown in Supplementary Fig. 21. We point out that although the photoresponse–and thus the absorption mechanisms–are mediated by interband and intraband transitions at telecom and sub-THz frequencies40,41,44, respectively, both exhibit cooling times on the order of a few picoseconds39,40,54. These results demonstrate that, despite having similar cooling times, the same capacitance and arrangement of source-drain electrodes, the sub-THz receiver exhibits a carrier-frequency-dependent bandwidth.

Fig. 4: Bandwidth-responsivity trade-off of receivers at different carrier frequencies.
Fig. 4: Bandwidth-responsivity trade-off of receivers at different carrier frequencies.
Full size image

a Bandwidth measurements of all investigated devices as a function of the carrier frequency, including the sub-THz range and the telecom frequency ( ~ 193.5 THz). Devices containing (not containing) a sub-THz antenna and back mirror are highlighted in a blue (red) background. The blue dashed lines indicate the optical bandwidth limitation range, whereas the purple dashed line indicates the bandwidth limitation of the PCB microstrip. b Bandwidth measurements as a function of responsivity in the sub-THz range of the graphene receivers. Devices that contain a sub-THz antenna and back mirror, and those that do not, are highlighted with green and blue circles, respectively.

We fabricate and test a 7th device (D7) that consists of a CVD graphene-based sub-THz receiver with a non-resonant split-gate architecture, that is, with a short split-gate length ( ~ 10 μm), a graphene channel of approximately the same length, and without a back mirror. In contrast to the other devices, Device 7 exhibits a bandwidth response of up to 40 GHz (limited by the Mach-Zehnder modulator, MZM) at a carrier frequency of 0.23 THz. Therefore, these results reinforce the optical bandwidth limitation imposed by the sub-THz cavity in previous devices and demonstrate that higher bandwidths can be achieved with graphene-based sub-THz receivers.

Next, we evaluate the bandwidth of the devices as a function of the measured responsivity in Fig. 4b. We observe that the sub-THz receivers with a responsivity higher than 1 mA/W, which contain the antenna and back mirror, show a limited 3 dB bandwidth ranging between ~1 and 2 GHz. Although Device 7 exhibits a very low responsivity of ~ 10 μA/W, its bandwidth reaches setup-limited values of 20–40 GHz. Our findings indicate that, although graphene intrinsically supports high bandwidths55,56, the optical coupling with the dipole antenna and back mirror–while enhancing responsivity within a specific spectral range–simultaneously imposes a limitation on the 3 dB bandwidth. Therefore, we anticipate that either employing broadband antennas, such as bow-tie or log-periodic designs–which provide broader resonance (i.e., a smaller Q-factor)–or reducing the size of the sub-THz cavity using polaritonic resonators45,57,58,59,60,61 can overcome the optical bandwidth limitation, as previously explained and detailed in Supplementary Note 12. By following these approaches, we could potentially achieve higher bandwidths and enable data rates exceeding 10 Gbits−1, while still maintaining high responsivity.

Discussion

In Table 1, we compare the performance of our developed receivers with those based on different technologies operating in the sub-THz range, as no previous graphene-based sub-THz data receivers have been reported. We acknowledge that these receivers do not necessarily pursue the same goals in terms of integration, size, and power constraints. The direct detection mechanism and CMOS-compatibility enable the graphene receiver to achieve low complexity and a minimal footprint, contributing to optimal performance. Thanks to this mechanism and the integration of a dipole antenna, the need for additional external components—such as a bulky horn antenna, optical amplifier, or optical receiver—is eliminated. In addition, graphene receivers enable the development of a Multiple-Input-Multiple-Output (MIMO) system15,62 due to their scalability, low complexity, and lower cost compared to other technologies.

Table 1 Quantitative comparison of photodetectors operating at sub-THz frequencies
Full size table

We point out that receivers based on other technologies may face issues such as incompatible high-temperature growth processes or epitaxial mismatch with silicon technology. Although III-V semiconductors can be transferred onto other platforms for heterogeneous integration63, this does not necessarily imply CMOS-compatibility. Several additional factors must be considered, including stringent contamination control and seamless integration within existing process flows. Notably, Ga and As degrade silicon device performance by introducing deep-level traps64,65,66. Given that the CMOS thermal budget is limited to approximately 400 °C, it is important to recognize that both Ga and As continue to exhibit diffusion under these conditions into Si, SiO2, and low-dielectric-constant materials. Similar concerns apply to the etching processes (wet or dry) required for GaAs, as well as the associated risk of cross-contamination in shared CMOS fabrication tools67.

Importantly, 2D material-assisted epitaxy is essential for the successful heterogeneous integration of III-V semiconductors63, highlighting the well-established scalability of 2D materials and their suitability for CMOS process flows. Moreover, carbon is not a blacklisted element, and graphene processing can be seamlessly incorporated into CMOS fabrication lines, as its dry etching relies on standard gases such as Ar and O2. Recent advances in large-scale graphene fabrication further strengthen its potential for CMOS-compatible technologies34,48,68,69,70,71,72. All the above-mentioned advantages position our graphene-based sub-THz receivers as a cost-effective solution that meets the size, weight, and power consumption (SWaP) requirements that future 6G technologies must satisfy.

An additional comparison with other graphene photodetectors in sub-THz range22,73,74,75,76,77,78,79,80,81,82,83,84,85 is presented in the Supplementary Table 4. Notably, none of them have demonstrated data stream detection. We stress that a higher data rate reception is potentially achievable by reporting a setup-limited bandwidth of 40 GHz for Device 7, for which it was not possible to measure the data stream detection owing to its low photosignal, responsivity, and high impedance. Recently, a bandwidth exceeding 400 GHz was demonstrated at telecom frequencies for a PTE-based graphene detector.55,56 Therefore, an intrinsically high bandwidth of graphene could be potentially achievable at sub-THz frequencies, suggesting the potential for data rates exceeding 500 Gbits−1. We further discuss the intrinsic performance limitations of graphene sub-THz receivers in Supplementary Note 4.

Future work

Our work demonstrates the potential of graphene receivers in the sub-THz range for wireless communications and paves the way for future research aimed at achieving even higher data rates. Other promising approaches to attain higher data rates include the use of advanced modulation schemes (e.g., four-level pulse-amplitude modulation), as illustrated in Table 1 for other technologies, and the development of a MIMO system that consists of multiple transmitting/receiving elements15,62, which are used in 5G wireless systems. The suggested MIMO receiver module includes sub-arrays, each of which is linked to a spatial channel containing a graphene receiver. The data rate of each individual graphene receiver would scale by the number of channels in the receiver module.

The previously mentioned implementation of arrays of graphene receivers can also solve the limitation of responsivity, particularly for CVD graphene-based devices, for measurements at larger distances with respect to the transmitter. Following this approach, the receivers can capture a larger portion of the incident sub-THz radiation. This high photoresponse could allow graphene-based receivers to operate without a voltage amplifier, resulting in a significant reduction in power consumption.

Our future investigations will adopt novel CVD growth methodologies to obtain high mobility values between  ~ 20,000 and 30,000 cm2/Vs in scalable graphene86,87. In fact, large-scale graphene has demonstrated these mobility values at room temperature and without the use of any encapsulant86. This high mobility in scalable graphene is expected to yield enhanced photodetection performance, following the trend shown in Fig. 3a. We also note that, for the PTE mechanism, optimal performance occurs at graphene doping levels near the charge neutrality point37,41,44,88, where the mobility is intrinsically high, as it depends on the doping level. All of the above-mentioned factors enable graphene receivers to play a key role in future 6G wireless communications.

Methods

Fabrication of the Graphene Receivers

Single Crystalline (SC) graphene growth and transfer

The growth process was carried out on a 35-μm thick copper (Cu) substrate using an Aixtron Black Magic cold-wall chemical vapor deposition (CVD) system. Prior to growth, the copper foil was electropolished to minimize surface roughness. The growth process began with an annealing step, during which the temperature was increased to 1000 °C in an argon atmosphere for 30 min. Subsequently, the precursors, 50 sccm of H2 and 30 sccm of CH4 (0.01% CH4 in Ar), were introduced to initiate nucleation. Growth was maintained for approximately 6 h at a constant pressure of 25 mbar. Subsequently, the CVD graphene was removed from the furnace and spin-coated with an A4-950K poly(methyl methacrylate) (PMMA) polymer from Microchemicals GmbH, which was subsequently cured at room temperature. The PMMA/SC-graphene/Cu composite was then transferred to 3 wt. % ammonium persulfate solution to etch away the copper foil. The PMMA/SC-graphene was rinsed by transferring it to deionized water and subsequently transferred onto a high-resistivity Si substrate. After drying at room temperature for one day, acetone/IPA baths were used to remove the PMMA layer from the top.

hBN encapsulated exfoliated graphene receivers

Graphene and hBN flakes are obtained through mechanical exfoliation from bulk highly oriented pyrolytic graphite (HOPG) and high-pressure, high-temperature-grown hBN single crystals, respectively, onto a 285 nm SiO2/Si++ substrate. Following the identification of the appropriate thickness using atomic force microscopy (AFM), the top hBN (T-hBN) flake was picked using polycarbonate (PC) (5 wt. % in chloroform)/PDMS/glass stamp at 100 °C. Subsequently, graphene and bottom hBN (B-hBN) were sequentially picked. The hBN-Gr-hBN stack/PC was then dropped at 180 °C onto an SiO2/ highly resistive Si double side polished substrate with prepatterned Cr/Au markers. Subsequently, PC was dissolved in chloroform for 30 min. One-dimensional source-drain electrodes were patterned using Elphy F50 Electron Beam Lithography (EBL) and Oxford Reactive Ion Etching (RIE) with SF6 and O2 gases. Subsequently, the metal deposition of Cr/Au was performed. The thickness of the metal electrode precisely matched that of the top hBN layer, and an additional dry etching step was introduced to isolate the photoactive channel. These procedures were conducted using 950 K-PMMA (from Microchemicals GmbH resist) as a lithographic mask. Subsequently, the sample was immersed overnight in acetone to facilitate a thorough clean lift-off process. Following this, another hBN flake (D-hBN) was dropped on top of the device, functioning as a dielectric layer that isolates the 5/200 nm thick (Cr/Au) dipole antenna. The dipole antenna was fabricated using Electron Beam Lithography (EBL) with a gap of 400–500 nm. The fabrication procedure for the TMD-Gr-hBN receiver described in this study remains consistent, with the exception of substituting the bottom hBN (B-hBN) layer with mechanically exfoliated WSe2. WSe2 flakes were exfoliated from the bulk and grown via chemical vapor transport (CVT).

CVD graphene receivers

Electron Beam Lithography (EBL) was used to pattern one-dimensional source and drain contacts, employing a 950 K PMMA mask on top of the SC CVD graphene on a highly resistive Si double side polished substrate. Subsequently, Reactive Ion Etching (RIE) with O2/Ar was employed to etch graphene. We implement an H-shaped graphene channel to improve the responsivity,37 and to reduce the contact resistance which is significant for CVD graphene-based receivers. A deposition of 5/20 nm of Cr/Au was performed, followed by overnight immersion in acetone to ensure an effective lift-off. Graphene undergoes a secondary patterning process using Electron Beam Lithography (EBL), followed by Reactive Ion Etching (RIE) to define the channel of the photodetector. A 40 nm-thick layer of alumina (Al2O3), which served as the dielectric and protective coating, was deposited at 250 °C via Atomic Layer Deposition (ALD) (Cambridge Nanotech). Subsequently, an EBL patterning process was used to pattern a dipole antenna with a  ~ 400–500 nm gap, followed by the deposition of a 5/200 nm thick Cr/Au layer. The same procedure was employed for fabricating the CVD-graphene/hBN receivers, with ALD-grown alumina substituted with exfoliated D-hBN. D-hBN was picked and dropped using a PC/PDMS/glass stamp at temperatures of 100 and 180 °C.

Electrical measurements

Electrical measurements were conducted at room temperature using a probe station from MPI Corporation. We use the probe station to check the devices at each fabrication step to verify the device integrity. The two-terminal resistance was measured by applying a 5 mV bias using a DAC, while the gate voltages for tuning the split gates were supplied by a Keithley 2640-B model.

Optical simulations

Optical calculations were performed using the full-vector 3D finite-difference time-domain (FDTD) method with Lumerical software. The computational cell dimensions are 8 × 8 × 12.6 mm3 with perfectly matched layer (PML) conditions applied to all boundaries. In the region occupied by the device, we employ a refined grid with 40 nm and 5 μm in the lateral x and y dimensions, respectively, and 10 nm in the vertical, z, direction. The sub-THz source is modeled as an incident plane wave reaching the device through an open aperture of 7.5 × 7.5 mm2 with a spectral range of 40–400 GHz. The dimensions of the device layers are described in the SI and Fabrication section in Methods.

Responsivity and NEP calculation

The external responsivity is given by: Responsivity = (IPTE/Pin) × (Afocus/Adiff)37,41,43,73, where Pin is the power measured by the commercial Erickson PM5B power meter from Virginia Diodes, Inc. The measured Pin is approximately −1.65 dB (0.68 mW) at 249 GHz, with the graphene receiver (or power meter) placed 15 cm from the horn antenna output. Afocus is the experimental beam area at the measured wavelength and Adiff is the diffraction-limited spot size. We measure the photocurrent IPTE from the output signal of the lock-in amplifier VLIA considering \({I}_{{{{\rm{PTE}}}}}=\frac{2\pi \sqrt{2}}{4\xi }{V}_{{{{\rm{LIA}}}}}\)37,41,43,73, where ξ is the gain factor in V/A (given by the lock-in amplifier). We use the ratio Adiff/Afocus for estimating the power reaching our photodetector since Adiff is the most reasonable value one can attain when considering the detector together with an optimized focusing system (e.g., using hemispherical lens) and it is widely used in the literature for comparing the performances among photodetectors37,41,43,73. This ratio is given by \({A}_{{{{\rm{diff}}}}}/{A}_{{{{\rm{focus}}}}}=\frac{{w}_{{{{\rm{0,diff}}}}}^{2}}{{w}_{{{{\rm{0,x}}}}}{w}_{{{{\rm{0,y}}}}}}\). In order to obtain w0,x and w0,y we use our experimental observation that the photocurrent is linear in laser power and measure the photocurrent while scanning the device in the x- and y-direction. Consequently, the photocurrent is described by Gaussian distributions \(\propto {{{{\rm{e}}}}}^{-2{x}^{2}/{w}_{{{{\rm{0,x}}}}}^{2}}\) and \(\propto {{{{\rm{e}}}}}^{-2{y}^{2}/{w}_{{{{\rm{0,y}}}}}^{2}}\), where w0,x and w0,y are the respectively obtained spot sizes (related to the standard deviation via σ = w0/2 and to the FWHM = \(\sqrt{2ln(2)}{w}_{0}\))37,41,43,73. We usually achieve w0,x = 10.1 mm and w0,y = 8.8 mm at λ = 1.4 mm (0.213 THz). We note that for the measurements performed to determine the responsivity, we did not employ focusing elements, such as lenses. Therefore, the sub-THz beam emitted from the horn antenna toward the graphene receiver, placed at a distance of 15 cm, is highly divergent. For the diffraction-limited spot, we consider \({w}_{{{{\rm{0,diff}}}}}=\frac{\lambda }{\pi }\), with λ the sub-THz wavelength. The diffraction-limited area is hence taken as \({A}_{{{{\rm{diff}}}}}=\pi {w}_{{{{\rm{0,diff}}}}}^{2}={\lambda }^{2}/\pi\).

Additionally, the noise-equivalent power (NEP) that characterizes the sensitivity of the photodetector is defined as NEP  = Inoise/Responsivity. Considering that our unbiased photodetector exhibits zero dark current, the noise current is therefore limited by Johnson-Nyquist contribution37,43. We use a noise spectral density \({I}_{{{{\rm{noise}}}}}=\sqrt{\frac{4{k}_{{{{\rm{B}}}}}T}{{R}_{{{{\rm{D}}}}}}}\), where kB corresponds to the Boltzmann constant, T is the operation temperature (300 K) and RD the device resistance. The calculated noise spectral density is used here because the experimentally measured value in our unoptimized setup is two orders of magnitude higher, as shown in Supplementary Figs. 26 and 27. This approach, also adopted in previous studies (see Supplementary Table 4), allows comparison of the intrinsic performance and obtained NEP of our device with those reported in the literature, independent of the noise from the electronics in the measurement setup43,49,88. Both the Johnson-noise-limited and experimentally measured NEP values are reported in the main text and Supplementary Note 9.