Research on a microring-assisted Mach-Zehnder modulator and its application in ONN computation

Electronic integrated circuits are struggling to meet growing computational demands. Optical neural networks are considered a promising path to overcome computational bottlenecks. In this paper, a computational structure is proposed. By integrating a microring resonator (MRR) with the Mach-Zehnder modulator (MZM) and applying wavelength division multiplexing (WDM) technology, the computational density is improved. Firstly, a theoretical model of microring-assisted MZM was established based on the coupling matrix. Using this model, the relationship between the weights and the modulation phase was achieved. The corresponding simulation model was built and validated to assess the effectiveness of the theoretical model. Finally, experimental verification was performed on the fabricated photonic integrated chip (PIC). The results showed that the computational structure proposed in this paper can perform the convolutional computation. The computational density reached 3.3 TOPS/TOPSmm²mm², and the power consumption required to modulate the smallest computing unit was 25.92 mW. This provides a practical technical solution for building high-efficiency optoelectronic hybrid computing systems.