
Nature News · Feb 18, 2026 · Collected from RSS
MainOptical telecommunication has profoundly accelerated the development of massive data interconnection3,5 and high-performance computing6,7. Despite great success in fibre communications, the future Internet of Everything demands high-throughput, wide-area coverage and low-latency signal delivery across all scenarios, as shown in Fig. 1, including high-speed fibre interconnections, multi-access wireless networks and hybrid fibre–wireless systems. To meet these demands, both fibre and wireless links require continuous improvement in single-lane speed and available bandwidth to enhance transmission capacity. Also, access architectures should migrate to the terahertz (THz) band and be deeply integrated with fibre technology, forming THz fibre–wireless networks to enable massive connectivity and low-cost signal processing8. Moreover, to cover areas beyond the reach of fibre infrastructure, such as remote habitats or harsh environments, wireless transmission links serving as hybrid relay nodes often require seamless fibre–wireless–fibre signal conversion, referred to as transparent relaying, to ensure low-latency and high-fidelity transmission across the interface1. Fig. 1: Integrated UWB all-optical telecommunication system.Conceptual drawings for all-optical ultra-broadband telecommunication connections (fibre–wireless congestion-free transmission, high-speed fibre interconnection and large-density fibre–wireless access) powered by the integrated photonics system. With the UWB integrated devices and seamless integration of the fibre–wireless system, high-throughput and low-latency all-optical telecommunication can be expected. OBPF, optical band-pass filter.Full size imageHowever, delivering comparable high-speed and low-latency data transmission ubiquitously across all scenarios remains a substantial challenge. One notable obstacle lies in the limited device operational bandwidth. Although baseband transceivers for fibre communications have already reached 100 GHz and beyond9,10,11, supporting transmission rates of 400 Gbps per lane12,13,14, such operation is already close to the bandwidth limit of present transceiver technology. In a unified fibre–wireless communication system, directly forwarding these wideband baseband signals to the wireless domain requires upconversion to the THz range (above 0.1 THz), which imposes greater challenges on the device broadband performance at the transmitter and receiver sides. Specifically, this requires a flat electro–optic–electro frequency response spanning several hundred GHz in the THz range, along with high saturation power in photodetection and high modulation efficiency, to achieve higher speed while preserving signal fidelity, which still remains difficult to meet. Although plasmonic modulators have been proposed to achieve near-THz bandwidth and support signal modulation deep into the THz range, they suffer from low modulation efficiency and high optical loss, which may limit the higher achievable lane speed15. On the other hand, uni-travelling carrier photodiodes (UTC-PDs) have become the dominant technique for chip-scale optical-THz signal generation16,17,18. Yet, relatively low OE bandwidth and limited sub-mW-level saturation power restrict the signal-to-noise ratio in THz applications.Another challenge concerns system architecture. Because electro–optic–electro signals conversions are usually processed at baseband in fibre communication and at intermediate frequency (IF) in wireless transmission, respectively, fibre–wireless systems usually require cross-band frequency mixing. To achieve this, both all-electric and hybrid optoelectronic approaches perform frequency mixing using frequency-multiplied electrical local oscillators (LOs)19,20,21,22,23,24. However, these approaches introduce electrical bandwidth constrains, further frequency-multiplication-induced noise accumulation and hardware complexity, which, in turn, limit system capacity and increase implementation cost. All-photonic-assisted wireless schemes have enabled direct wireless-to-optical signal conversion and subsequent processing in the optical domain, offering substantial hardware savings and excellent frequency conversion consistency2,25,26. Full link functionalities and broadband spectrum adaptability have also been demonstrated on an integrated photonics platform for all-optical wireless communications27. Nevertheless, the demonstrated single-lane data rates so far remain limited to less than 80 Gbps (ref. 25). Although various multiplexing techniques can further enhance overall capacity26, such aggregation introduces considerable complexity in signal encoding and decoding, thereby hindering the realization of low-latency transparent relaying. Furthermore, signal distortions caused by both linear and nonlinear impairments become increasingly pronounced as the signal bandwidth increases towards the 100-GHz range. This makes traditional linear digital signal processing (DSP) algorithms largely ineffective for broadband fibre–wireless convergence, especially as future infrastructure demands lane rates exceeding 400 Gbps.Here we present a unified UWB fibre–wireless communication solution based on integrated photonics. Using a pair of photonic EO and OE converters with state-of-the-art bandwidths exceeding 250 GHz, consisting of a thin-film lithium niobate (TFLN) modulator and a modified UTC-PD, we realize a fibre–wireless bandwidth-shared transmission scheme with more than 100 GHz channel bandwidth available in both the fibre and wireless links. Facilitated by efficient signal modulation, high-power photodetection and a unified complex-biGRU algorithm, our system achieves high-quality data transmission in both fibre and wireless links. The single-lane data rate is boosted to, to the best of our knowledge, the highest levels in both scenarios, with 512 Gbps achieved over the fibre link and 400 Gbps over the wireless link at the THz band. More practically, an 86-channel 8K real-time video transmission has been realized with 1-GHz channel bandwidth across 138–223 GHz, which is one order of magnitude larger than the standard 5G protocol. Our work paves the way towards the full integration of ultra-broadband all-optical communication systems and will be a promising route for next-generation telecommunications.Ultra-broadband on-chip EO and OE conversionThe ultra-broadband ability of our demonstration is ensured by a TFLN modulator and an indium phosphide (InP)-based UTC-PD, both of dedicated design and fabricated for large bandwidth and efficient conversion. The device photograph and cross-section schematic diagram of the integrated EO modulator are shown in Fig. 2a. The chip is fabricated on a 360-nm X-cut lithium niobate (LN) wafer, with a 500-µm-thick quartz substrate used to reduce microwave loss (Methods). To compensate for the slow wave effect induced by quartz substrate, a periodic capacitively loaded travelling-wave electrode (that is, slow-wave electrode) structure is applied28, with excellent velocity match and impedance match (Supplementary Note 1). The input and output pads are also appropriately designed to have a 50-Ω characteristic impedance as well as 50-Ω on-chip terminators at the end of the capacitively loaded travelling-wave electrodes.Fig. 2: Characterizations of fundamental UWB building blocks and short-reach interconnection.a, Optical image and 3D cross-section view of the TFLN modulator. b, Optical spectrum of the modulated sidebands (carrier omitted). The bandwidth measurement uncertainty from 110 to 220 GHz is ±0.4 dB owing to the power uncertainty of the OSA. c, Normalized EO response from 1 to 220 GHz. The solid line is the extrapolated response by fitting the experimental data. d–f, Optical image and 3D schematic diagram (d), RF output power (e) and OE bandwidth (f) of the modified UTC-PD. g, Schematic of the IMDD data transmission set-up. EA, electrical amplifier; PC, polarization controller. h, Eye diagrams of 210-Gbaud NRZ (left) and 196-Gbaud PAM-4 (right) captured by the DSO. i, BER for NRZ and PAM-4 transmission at different symbol rates using the complex-biGRU algorithm. All results are considered within the given HD-FEC threshold. j, Reconstructed eye diagrams and BERs of 256-Gbaud NRZ and PAM-4 signals.Full size imageWe then show the first >200 GHz 3-dB EO response characterization of a TFLN modulator. Including the lensed fibre-to-chip coupling loss (about 3.8 dB per facet), the total optical loss comes to about 8.2 dB, indicating an on-chip insertion loss of 0.6 ± 0.3 dB. The EO response is measured using a lightwave component analyser (LCA) under 110 GHz and optical spectrum analysis for higher frequencies (Methods). Figure 2b shows the sidebands of the modulated signal in the range 140–220 GHz. The inset depicts a flat optical spectrum over the test range with a ripple of 2.1 dB. The whole EO response shows an ultrahigh experimental bandwidth of >220 GHz (Fig. 2c) and an extrapolated result of about 260 GHz (with respect to 1 GHz). To the best of our knowledge, this is the highest experimentally measured EO bandwidth reported for TFLN modulators, excluding previously extrapolated or estimated results (Extended Data Table 1). Benefiting from the excellent EO bandwidth, radio frequency (RF) half-wave voltage Vπ,RF of 6.2 V at 200 GHz can be calculated (Methods). To ensure both fibre and wireless communications, the device is designed to achieve better balance between EO bandwidth and modulation efficiency while maintaining lower insertion loss. Compared with other EO bandwidth enhancement techniques15,29,30,31, our design also obtains a flat EO response with the maximum deviation of 0.5 dB above the 0 dB level. Such property offers substantial improvements towards faithful high-speed wireless communication requiring wide spectral occupancy and uniform signal integrity in multichannel real-time video transmission.The optical image and corresponding schematic diag