With the rapid development of high bit-rate wireless services driven by mobile internet, AI computing, high-definition videos, virtual reality/augmented reality (VAR) applications, and so on, the demand for wireless data rates has grown explosively in the past decades^1,2. Supporting such fast data rate at tens of Gbit/s pushes the carrier frequency to the THz (0.1–10 THz) domain in 6G to accommodate the bandwidth-hungry wireless services³⁻¹⁴. Recently, considerable efforts have been invested in developing photonic techniques targeting ultrafast data rate wireless communication at THz frequencies¹⁵⁻¹⁸.

Photodiodes (PDs) serve as a key component of photonic transmitter to realize optical-to-electrical conversion. In particular, uni-traveling-carrier photodiodes (UTC-PDs) have been developed to simultaneously achieve both ultrawide bandwidth and high output power¹⁹. For instance, flip-chip-bonded 6-μm-diameter UTC-PD has demonstrated bandwidth exceeding 110 GHz as well as radio frequency (RF) power of 7.8 dBm at 110 GHz²⁰. Flip-chip bonded modified UTC-PDs (MUTC-PDs) exhibiting a 3-dB bandwidth up to 120 GHz or a saturated RF output power of −2.6 dBm at 160 GHz have also been reported²¹.

Module packaging of high-speed PDs is important for their applications in millimeter wave or THz wireless communications. Packaging of GHz-bandwidth PDs is relatively mature. For example, fully packaged PD modules with coaxial connectors for below 30 GHz applications have been reported^22,23. However, at frequencies over 100 GHz, high transmission losses as well as parasitic modes and resonances due to the packaging structure will seriously degrade the performance of PD modules. By adopting a conductor-backed coplanar waveguide (CBCPW) to suppress mode-mismatch between the PD and the RF connector, a packaged PD with 1-mm coaxial connector was reported to exhibit a bandwidth over 90 GHz²⁴. A bandwidth of 145 GHz was estimated for a packaged PD equipped with 0.8-mm coaxial connector, but the frequency response exhibited a significant roll-off above 130 GHz²⁵.

For PD modules operating at higher frequencies, waveguide-integrated PD modules offer a promising solution for high-power THz output due to the low-loss nature of rectangular waveguides and the availability of a wide range of standard components. For instance, a WR-6 waveguide packaged UTC-PD module for D-band (110–170 GHz) operation was developed, with an output power around −5 dBm²⁶. A WR3-band (220–320 GHz) PD module containing an MUTC-PD with short-stub resonant matching circuit was reported to achieve a maximum output power of 134 μW at 264 GHz²⁷.

Currently, many researchers have employed commercial UTC-PDs to construct photonic-assisted THz wireless communication systems, in which THz signals can be generated by beating two lasers with different wavelengths at the transmitting end. A demonstration showcased a multichannel THz wireless transmission utilizing the quadrature phase shift keying (QPSK) modulation format at a carrier frequency of 200 GHz, achieving a data rate of up to 75 Gbps²⁸. In another experiment²⁹, a single-input single-output (SISO) wireless link was established at 237.5 GHz by utilizing a remote UTC-PD. Through the combination of modulation formats, such as QPSK and quadrature amplitude modulation (QAM), a maximum data rate of 100 Gbps was attained. Simultaneous photonic-assisted THz communication in three bands (150, 250, and 325 GHz) has also been demonstrated³⁰. The net data rates reached 59.813, 74.766, and 93.333 Gbps, respectively.

In this work, we report a WR-5 waveguide packaged PD module for THz communications. The module contains a 4-μm-diameter back-illuminated MUTC-PD with optimized epitaxy structure and coplaner waveguide (CPW) electrodes to ensure wide bandwidth and high saturation power. Passive circuits including bias-tee and E-plane probe are incorporated into the module. The packaged PD module has demonstrated a flat frequency response with fluctuations within ±2.75 dB in the entire G-band, and the saturated output power reaches −7.8 dBm at 140 GHz.

Photonics-assisted THz wireless communication experiment have been caried out. Thanks to the high saturation power and high conversion efficiency of the PD module, meter level short distance wireless data transmission has been demonstrated without additional power amplifiers after the optical-to-electrical conversion. To ensure high-performance signal transmission and avoid nonlinear effects caused by the MUTC-PD under high optical power, the 8- amplitude phase shift keying (APSK) modulation format is employed, and transmission rates of 75 and 90 Gbps over 1-m free space at 150.5 and 210.5 GHz carrier frequencies, respectively, are demonstrated.

The epitaxial structure of our MUTC-PD is shown in {L-End} Fig. 1(a), similar to the one reported in our previous work^{Fig. 31}. A gradient-doped absorption layer is employed to alleviate the space charge effect for improved saturation performance, while a cliff layer with optimized doping is adopted to sustain electron velocity overshoot for enhanced bandwidth. Inductive peaking can be implemented with high-impedance CPW to manipulate the frequency response, enabling a reduced power roll-off at the high frequencies of interest. As shown in {L-End} Fig. 1(b), a flat output power over 140–220 GHz can be maintained when the inductance is designed to 50 pH (5E-11H).

The 3D schematic of the 4-μm-diameter PD is illustrated in {L-End} Fig. 1(c), which adopts a triple-mesa structure for improved fabrication yield and consistency. The first mesa is etched by inductive coupled plasma (ICP) and stopped slightly below the p-n junction to load most of the electric field. The second mesa is defined by a combination of dry-etching and wet-etching process^{Fig. 31}. Diluted HCl is used to stop precisely at the InGaAs layer for ohmic contact formation. The bottom mesa is etched down to the semi-insulating InP substrate to provide insulation. Ti/Pt/Au and Ni/Au electrodes are formed on the p- and n-contact layers by magnetron sputtering. Scanning electron microscope (SEM) image of the fabricated device is shown in {L-End} Fig. 1(d). Due to the relatively complex fabrication process, the performance of PDs on the same wafer may vary somewhat.

The frequency response of the fabricated 4-μm-diameter MUTC-PD at G-band are measured with a two-laser heterodyne system. The frequency responses under different photocurrents with a fixed reverse bias of 2 V are plotted in {L-End} Fig. 1(e). The PD exhibits ultra-flat frequency responses over the entire G-band under various photocurrents, together with a responsivity of 0.07 A/W. The saturation characteristics of the PD is not measured, as the PD performance might deteriorate after the high-photocurrent saturation measurement, which will affect the evaluation of the frequency responses of the PD after packaging.

To package the high-speed PD chip into a compact module with waveguide output, it is essential to realize smooth transition from the CPW electrodes on the PD to the waveguide port. In addition, a bias-tee is also required to provide dc bias to the module. These are implemented by passive circuits formed on 127-μm-thick quartz substrate, which is adopted for reduced dielectric loss at high frequencies. The configuration of the passive circuits is illustrated in {L-End} Fig. 2(a).

A novel bias-tee is proposed and fabricated based on interdigital structure coplanar waveguide, securing broadband and low loss transmission at G-band compared with the commonly reported bias-tee structures³²⁻³⁴. The bias-tee consists of an RF-choke and a DC-block. The RF-choke is formed by an interdigital inductor³⁵, which serves as a quarter wavelength transformer, resulting in a virtual RF open circuit. Two open circuit stubs in parallel are employed to improve the RF signal isolation³⁶. The DC-block is formed by an interdigital capacitor³⁵ to ensure low loss and smooth transition over wideband, whereas traditional coupled-line based DC-block exhibits bandpass performance³⁴. The bias-tee performance is test by microwave probes via a back-to-back configuration, as shown in {L-End} Fig. 2(b). The measured insertion loss (IL) is less than 4 dB, while the return loss (RL) is about 10 dB. The measurements are in reasonable agreement with the simulation results.

In previous studies, the planar transmission line to rectangular waveguide transition at D-band employing wire bonding probe³⁷ or metal ridge³⁸ suffers from poor repeatability and large loss. The transition implemented by a stepped-impedance E-plane probe converts the quasi-TEM mode of the microstrip to the dominant TE₁₀ mode in the waveguide with low loss. Smooth transition over the G-band is realized through impedance matching by varying the width of microstrip in three steps. A back-to-back package is fabricated to verify the performance of the E-plane probe. As plotted in {L-End} Fig. 2(c), an RL over 10 dB is recorded over the G band, while the measured IL is around 2.25 dB, corresponding to an IL of 1.125 dB for a single microstrip-to-waveguide transition. Again, the measurement results are in fair agreement with the simulations.

The schematic of the PD module with WR-5 waveguide output is shown in {L-End} Fig. 3(a). The PD chip is flip-chip bonded to the passive circuits including bias-tee, E-plane probe and DC components. An electrostatic discharge (ESD) chip is included in the DC path, so as to protect the chip against static electric damage. A lensed fiber is used to couple light into the PD, and the converted THz signal is extracted through a WR-5 waveguide.

During the package, planar passive circuits, including the bias-tee and the E-plane probe, are first fixed onto the bottom block of the module by silver-filled epoxy. Gold-wire bonding is employed for chip-level interconnections, as shown in {L-End} Fig. 3(b). The length and height of the bonding wires are kept at a minimum so as to suppress parasitic inductance.

Next, the PD chip is flip-chip bonded to the passive circuits via gold bumps formed at the end of the bias-tee, as shown in {L-End} Fig. 3(c). The flip-chip bonding conditions is optimized to ensure that no degradation in PD performances occur during bonding. As plotted in {L-End} Fig. 3(d), the I-V characteristics, including dark current and contact resistance remain almost unvaried after the flip-chip bonding. As the final step, a lensed fiber is attached to the PD module to ensure best responsivity.

The fully packaged MUTC-PD with standard WR-5 output is shown in {L-End} Fig. 4(a). The frequency responses tested under a reverse bias of 2 V are plotted in {L-End} Fig. 4(b). The PD module exhibits a flat frequency response with a fluctuation within ±2.75 dB over 140–220 GHz under various photocurrents. Compared with {L-End} Fig. 1(e), the packaged PD shows a power drop by about 1.09 dB, which is consistent with the insertion losses of the bias-tee and the probe, indicating that no additional losses are introduced during the packaging process. The frequency-dependent saturation characteristics of the packaged PD is shown in {L-End} Fig. 4(c). The saturation photocurrent under 140, 150, and 160 GHz is 8.6, 8.Fig. 3, 7.7 mA, corresponding to a saturation power of −7.8, −8.Fig. 3, and −8.9 dBm, respectively. {L-End} Fig. 5 summarizes the saturation output power versus the operation frequency of UTC-PD modules reported in the literature. The proposed MUTC-PD module with rectangular waveguide output shows high saturation power at high frequencies without employing resonant matching circuits.

The MUTC-PD has great potential for applications in photonic-assisted THz wireless communications. The flat response over the G-band enables easy frequency agility at the transmitting end, making it suitable for diverse application scenarios. Experiments are carried out to showcase the ultra-broad bandwidth MUTC-PD module in a photonic-assisted THz high-speed wireless communication system.

{L-End} Fig. 6 provides an overview of the photonic-assisted THz high-speed wireless communication system and the offline digital signal processing (DSP). The first step in the transmission process involves generating and mapping the data sequence according to regular 8-APSK. The baseband signal then undergoes up-sampling and pulse-shaping using a root-raised cosine filter with a roll-off factor of 0.1. The resulting baseband signal, depicted in {L-End} Fig. 6(a), is then sent to an arbitrary waveform generator (AWG) with a sampling rate of 64 GSa/s. Subsequently, the signal is modulated onto the light emitted by an external cavity laser (ECL) using an IQ modulator. ECL-1, operating at Fig. 19Fig. 3.Fig. 3985 THz with a linewidth of 100 kHz, serves as the light source for this system. Finally, the signal light is amplified by an erbium-doped fiber amplifier (EDFA) and aligned with the polarization state of the local oscillator (LO) light emitted by ECL-2 through a polarization controller.

In the experiment, the frequencies of the light emitted by ECL-2 are set to 193.549 and 193.609 THz, respectively, to verify the ultra-broadband response of the MUTC-PD module and its application in frequency agile THz communication systems. The optical spectra recorded after the optical coupler (OC) are shown in {L-End} Fig. 6(b) and {L-End} Fig. 6(c), revealing a frequency difference of 150.5 and 210.5 GHz between the two lasers, respectively. Subsequently, the light coming out of the OC is amplified by another EDFA and fed into the MUTC-PD module. The generated THz signal is then launched by the horn antenna (HA). The high saturation power and high conversion efficiency of the PD module enables meter-level short distance THz communication transmission without the need for additional power amplifiers. For effective transmission of THz signals at different frequencies, HAs with different center frequencies are employed at the transmitting end.

At the receiving end, receivers at different frequency bands are adopted to receive THz signals at 150 and 210 GHz, respectively. After 1-m wireless link transmission, the transmitted THz signal is captured by the HA, amplified by a low noise amplifier (LNA), and then mixed with the LO signal generated by frequency multiplier. The signal is down-converted to the intermediate frequency (IF) after passing through the mixer. The IF signal is then amplified by an electrical amplifier (EA) and captured by an oscilloscope (OSC) at a sampling rate of 80 GSa/s. Subsequently, the captured signal is processed using an offline DSP to recover the original data. The offline DSP mainly includes down-conversion, matched filtering, down-sampling, and some traditional coherent communication DSP, as shown in {L-End} Fig. 6. As the first step, the Gram-Schmidt orthogonal projection (GSOP) is adopted to correct the I/Q imbalance, followed by clock recovery based on the fast square-timing-recovery algorithm. The signal is recovered by blind equalization based on the constant modulus algorithm (CMA). Following the frequency offset and carrier phase estimation (FOE & CPE), a decision-directed least mean square (DD-LMS) algorithm is employed to further enhance the system performance. Finally, the signal undergoes APSK demodulation, and the error vector magnitude (EVM) is calculated.

The modulation format adopted in the experiment is 8-APSK, which offers improved frequency spectral efficiency and achieves a higher data rate compared with common low-order modulation formats, such as OOK or QPSK. Furthermore, the 8-APSK modulation scheme demonstrates superior resistance to nonlinear distortion compared to the standard 8-QAM modulation scheme. {L-End} Fig. 7 shows the EVM performance of the 8-APSK signals transmitted over a 1-m wireless distance at different baud rates at the center carrier frequencies of 150.5 GHz and 210.5 GHz, respectively. In our experiment, bit-error-free transmission of 5-GHz baud rate 8-APSK signals are verified at both 150.5 GHz and 210.5 GHz. At 150.5 GHz carrier frequency, the system can support transmission at 25-GHz baud rate below the forward-error-correction (FEC) threshold of 1E-2, corresponding to a transmission rate of 75 Gb/s. The bit-error-rate (BER) value is 9.66E-Fig. 3 and the EVM value is 26.92%. Similarly, over Fig. 30-GHz baud rate transmission can be achieved at 210.5 GHz, with the EVM value of 26.71% and BER value of 9.06E-Fig. 3. In the experiment, we also endeavored to attain higher data rates by employing a high-order QAM scheme. Specifically, we conducted the THz wireless communication experiment utilizing a 20-GHz baud rate standard 16-QAM signal at a carrier frequency of 150.5 GHz. However, the tested BER of Fig. 3.7E-2 significantly exceeded the acceptable threshold of 1E-2. The modulation orders at this bandwidth are limited by the over-all system signal-to-noise ratio as well as the nonlinearity impairment. In {L-End} Fig. 7, insets (i–iv) depict the signal constellation points for 5-GHz baud rate and the largest transmission baud rate at 150.5-GHz and 210.5-GHz, respectively.

As revealed by {L-End} Fig. 7, the packaged PD module shows great potential in ultra-broadband and high-speed THz communication in G-band. It is worth mentioning that thanks to the high output power of the PD, high-speed transmission of 90 Gb/s is achieved without any power amplifier after the PD. This helps reduce the complexity and costs of the system, as well as maintaining a high signal-to-noise ratio (SNR). In our future research, we plan to perform thorough testing at different distances to gain a comprehensive understanding of the performance attributes of our MUTC-PD, so as to facilitate the practical application of our system in real-world scenarios.

The baud rate bandwidth demonstrated in the experiment is mainly limited by the frequency range of the electrical devices in the system. By employing electrical devices with wider bandwidth and advanced DSP algorithm such as deep learning-based end-to-end autoencoder methods³⁹, it is believed that the PD module can support THz signal transmission with even larger bandwidth and higher rate.

We have demonstrated a compact MUTC photodiode module with WR-5 waveguide output. The PD module exhibits a flat response in the frequency range of 140–220 GHz, and is successfully employed for photonic-assisted THz wireless communications. Transmission rates of 75 and 90 Gbps over 1-m free space are demonstrated at 150.5 and 210.5 GHz carrier frequencies, respectively. The over 80 GHz bandwidth of the PD module at G-band promises other broadband applications, such as THz spectroscopy and imaging. Furthermore, with the high saturated output power, the PD module has the potential to achieve high-performance 6G THz wireless communication. In our future research endeavors, we aim to delve deeper into the application capabilities of the MUTC-PD, encompassing high-speed wireless bi-directional communication. Additionally, we envision a framework where both the MZM and PD modules can be seamlessly integrated on-chip, thereby facilitating chip-level wireless data transmission with integrated antennas. This integrated approach holds significant promise for realizing efficient, high-speed, and compact wireless communication solutions across a wide range of applications.

Furthermore, metasurfaces, renowned for their precise control over electromagnetic wave properties, offer the possibility of engineering specific responses to decouple and independently control different wavelengths, incident angles, spin angular momentums (SAMs), and orbital angular momentums (OAMs). This capability holds significant promise for high-capacity THz communications. Recent advancements in metasurface research⁴⁰⁻⁴³, including novel designs and fabrication techniques, enhance their potential applications in THz communication systems. By integrating metasurfaces into THz devices, alongside well-designed PD modules, such as antennas and modulators, we can achieve enhanced beamforming, signal modulation, and processing capabilities. This integration promises to significantly increase communication throughput, reliability, and efficiency in THz wireless networks, paving the way for future high-speed data transmission and communication technologies.

This work was supported in part by National Key Research and Development Program of China (No. 2022YFB2803002); National Natural Science Foundation of China (Nos.62235005, 62127814, 62225405, 61975093, 61927811, 61991443, 61925104 and 61974080); and Collaborative Innovation Centre of Solid-State Lighting and Energy-Saving Electronics.

The authors declare no competing financial interests.