Weizhen Yu

Wavelength-division multiplexing (WDM)-compatible mode-division multiplexing (MDM) approach is a promising optical communication technique that leverages multiple spatial modes on different wavelengths within an optical waveguide to boost the data capacity. In a WDM-compatible MDM transmission system, a wavelength-selective mode exchanger is a key component which can selectively fulfill the mode conversion on a specific wavelength of the light. Here we propose a compact and reconfigurable wavelength-selective mode exchanger on a silicon photonic chip. The device employs a symmetric Mach–Zehnder interferometer (MZI) structure, with optical mode couplers at both input and output for mode conversion. Each MZI arm features a single-mode rib waveguide, integrated with multiple ultra-compact microdisk resonators (MDRs) for wavelength selectivity, and a metallic micro-heater on each MDR for tuning. By adjusting the micro-heaters, a controlled phase-shift difference between the two arms enables wavelength-selective mode conversion. The mode conversion bandwidth is determined by the MDR resonance bandwidth, and can be enhanced by aligning the resonances of multiple MDRs. Experimental results demonstrate successful transverse-electric (TE0 and TE1) mode conversions across three wavelength channels. The device exhibits strong reconfigurability, allowing tunability of the central wavelength, channel spacing, and bandwidth. Using the chip in an optical transmission system, WDM-compatible MDM transmission is demonstrated. The proposed on-chip wavelength selective mode exchanger holds key advantage of scalability since the WDM channel can be hugely increased by adding more ultra-compact MDRs, which paves the way for future large-scale WDM-compatible MDM optical communication systems.

We propose a photonic approach for generation of independently tunable dual-band linearly-chirped microwave waveforms (LCMWs) based on a Fourier-domain mode-locked opto-electronic oscillator (FDML-OEO). The FDML-OEO is implemented with the use of an independently-tunable dual-passband on-chip microwave photonic filter (MPF) realized by phase modulation to intensity modulation (PM-IM) conversion. In the integrated PM-IM implementation, an optical filter with a tunable dual-passband is a key component that incorporates two separated racetrack microring resonators (MRRs) with an ultra-narrow bandwidth. A metallic micro-heater is placed on top of each MRR for resonance wavelength tuning when a driving signal is applied. When the periods of the two applied driving signals match the round-trip time of the OEO loop simultaneously, Fourier-domain mode locking is reached, and dual-band LCMWs can be generated. Thanks to independently tuning of each MRR, each sub-band of the dual-band LCMW generated by the FDML-OEO is independently tunable in terms of central frequency, signal bandwidth and chirp rate. A silicon-based ultrahigh-Q dual-bandpass optical filter chip is fabricated and characterized. With the use of the fabricated chip, the generation of dual-band LCMWs is demonstrated, of which each sub-band can be tuned in the central frequency from 7.5 to 12.5 GHz, the bandwidth from 0.5 to 1.5 GHz, and the chirp rate with positive or negative value. The proposed integrated independently tunable dual-band LCMW generation system features great advantage of waveform tunability, which is potential to be widely used in multi-band multi-functional radar systems applications.

The realization of optical nonlinear activation functions (NAFs) is essential for integrated optical neural networks (ONNs). Here, we propose and experimentally demonstrate a photonic method to implement reconfigurable and low-threshold all-optical NAFs based on a compact and high-Q add-drop microring resonator (MRR) on silicon. In the experiment, four different NAFs including softplus, radial basis, clamped ReLU, and sigmoid functions are realized by exploiting the thermo-optical (TO) effect of the MRR. The threshold to implement NAFs is as low as 0.08 mW. As a demonstration, a handwritten digit classification benchmark task is simulated based on a convolutional neural network (CNN) using the obtained activation functions, where an accuracy of 98% is realized. Thanks to the unique advantages of ultra-compact footprint and ultralow threshold, the proposed nonlinear unit is promising to be widely used in large-scale integrated ONNs.