Chinese researchers have achieved a groundbreaking advancement in optical data transmission with the development of a silicon photonic integrated high-order mode multiplexer chip. This innovation, led by a research team from Fudan University, enables ultra-high-capacity on-chip optical data transmission, providing a transformative solution for high-performance computing and artificial intelligence (AI) applications, Global Times reported on Thursday.
The newly developed chip, published in Nature Communications, addresses critical challenges in large-scale AI model training, data center connectivity, and high-performance computing. As AI models expand, the need for higher communication bandwidth between intelligent computing chips has intensified. Fudan University’s team tackled this challenge by integrating multi-dimensional multiplexing technology into on-chip optical interconnects, optimizing design and significantly enhancing transmission efficiency.
According to Science and Technology Daily, the chip demonstrates exceptional power consumption efficiency and minimal transmission latency while achieving an unprecedented data transmission throughput. Tests reveal that the chip supports a staggering 38 terabits per second (Tbps), capable of transferring 4.75 trillion model parameters per second. This breakthrough not only improves communication efficiency in large-scale computing clusters but also lays the groundwork for AI, model training, and GPU-accelerated computing.
As the global tech industry explores alternatives to traditional electronic transmission, photonic chips have garnered increasing interest. While electronic chips remain dominant, the shift toward optical communication is gaining momentum, experts say. Veteran telecom industry observer Ma Jihua emphasized the significance of this development, stating that integrating photonic and electronic components has been a longstanding challenge.
“The integration of photonics-based interconnection significantly reduces transmission latency and enhances efficiency,” Ma told Global Times on Thursday. He noted that high-order multiplexing provides superior transmission efficiency compared to traditional single-mode systems, making it particularly valuable in AI applications that demand high-speed, high-volume data transfer.
Beyond improving transmission speeds, the technology has strong scalability and compatibility with various high-performance computing applications. Analysts believe it will play a crucial role in advancing AI model training and large-scale parallel computing. Moreover, the development could give China a significant competitive edge in next-generation networking and semiconductor research.
Ma further highlighted that the innovation could substantially boost domestic AI training capabilities, improving power consumption efficiency and making large-scale AI model deployment more commercially viable. “Major breakthroughs in application may come within the next three to five years,” he predicted.
This latest achievement reflects China’s commitment to advancing semiconductor and photonic chip technology, reinforcing its position as a global leader in AI and high-performance computing infrastructure. With increasing demand for high-speed data processing, such innovations are expected to shape the future of AI and data center efficiency worldwide.
