Recently, Researcher Zhang Junwen and Professor Chi Nan from the School of Information Science and Engineering at Fudan University have collaborated with relevant research teams to introduce multi-dimensional multiplexing technology into on-chip optical interconnect architecture through precise design and optimization. This not only significantly improves the data transmission throughput, but also excels in power consumption and latency. It has strong scalability and compatibility and is suitable for a variety of high-performance computing scenarios. On this basis, the team designed and developed a silicon photonics integrated high-order mode multiplexer chip, realizing ultra-large capacity on-chip optical data transmission. Experimental results show that the chip can support a data transmission speed of 38Tb per second, which means that 4.75 trillion parameters of a large model can be transferred in 1 second in the future. This significantly improves the communication performance and reliability between large model training and computing clusters, and provides strong support for applications such as artificial intelligence, large model training, and GPU accelerated computing.