As the scale of AI models increases dramatically, the demand for computing power and efficiency also increases rapidly.Note) bottlenecks, and the reality that commercialization of quantum computing remains a long way off. Photonic computing, with its advantages of low energy consumption, high bandwidth, and zero heat loss, is gradually moving towards industrialization in the global market. For Taiwan, which relies on semiconductor manufacturing as its core, this path may become a significant window of opportunity for renewed innovation.
Note:The von Neumann architecture was proposed by John von Neumann, an American mathematician known as the "Father of Game Theory" and the "Father of Computers." It is a computer design concept that stores program instructions and data in the same memory space. It has become the foundation of modern computer design and is also the Princeton architecture.
Taiwan's Strategic Axis: From Mass Production to Optoelectronic Integration
Taiwan's AI and semiconductor policies have clearly identified silicon photonics as a key strategic direction. The Taiwanese government has explicitly included silicon photonics technology in its "Ten Major AI Infrastructure Plans," projecting it to drive related output to over NT$5100 billion. This demonstrates that photonic computing is no longer simply a technical issue; it is now a core pillar of Taiwan's national development.
On a technical level, TSMC is a leader in silicon photonics. Its "COUPE Silicon Photonic Engine" combines advanced packaging with optical modules, opening up the potential for co-packaged optics (CPO) technology. Furthermore, TSMC has nearly doubled Intel's silicon photonics patent applications in the United States, highlighting its leading position in optoelectronics technology.
On the other hand, NVIDIA is also preparing to invest in this field and plans to incorporate silicon photonics into its high-performance AI switches, which means that silicon photonics will become a key technology for important developments in the future.
Photonic computing becomes a breakthrough point for AI data centers
Photonic computing architectures encompass diverse approaches, including free-space optics, photonic chips, and fiber-optic computing systems. They offer potential for breakthroughs in data transmission and energy efficiency, particularly for AI training and inference. As for photonic chips, if successfully integrated with electronic chips in the future, they will become a key driver for AI model deployment and the development of smart data centers.
Taiwan already has a solid foundation in packaging, 3D IC, and silicon photonics integration, and the commercial demand for AI models is driving the early adoption of photonic computing technology. For example, NVIDIA plans to use silicon photonics high-speed interconnect technology to increase AI switch speeds to 1.6Tbps, potentially significantly improving AI data center performance.
Still facing multiple challenges: integration, cost and packaging to turn the tide
Currently, photonic computing still faces several challenges, including stability, accuracy, optical-to-electrical conversion latency, and data storage and packaging integration. Nevertheless, hybrid packaging technologies, such as co-packaged optics (CPO), offer the most promising path for implementation in the short term. In the medium term, hybrid optical-electrical computing architectures integrating time and space may become the next step in the computing efficiency revolution.
For Taiwanese companies, leveraging their existing manufacturing and packaging advantages, coupled with government policy support and the resonance of AI model demand, they will have the opportunity to take the lead in this "optical task revolution" and reshape their position in the global semiconductor value chain.
