GF SEC: The turning point of silicon photonics accelerating penetration has arrived. We continue to recommend paying attention to the CPO and OIO industry chain.

GF SEC released a research report stating that silicon photonics technology has become an important pathway for the "More Than Moore" era after Moore's Law, which is a clear technological development trend. It is recommended to focus on manufacturers with silicon photonics chip design capabilities and mass production capabilities of silicon photonics modules. At the same time, in the division of labor in the CPO and OIO industry chains, it is suggested to continue focusing on: (1) traditional optical module manufacturers with a deep understanding of optoelectronic packaging and relevant technical reserves; (2) companies with abundant technical reserves in passive connection technologies such as MPO/MTP/MMC, FAU, and high-quality customer resources; (3) manufacturers with mass production capabilities of CW high-power laser chip; (4) optical testing and coupling equipment manufacturers. GF SEC's main points are as follows: Silicon photonics can solve the core contradiction of the increasing demand for computing power in the AI/ML era and the failure of Moore's Law. In the process of optical signal transmission, less attenuation and higher transmission bandwidth, silicon-based optoelectronics have advantages such as large bandwidth, high speed, low energy consumption, and strong interference resistance. As Moore's Law gradually becomes ineffective and the traditional electronic chip transmission rate encounters bottlenecks, silicon photonics technology can adapt to the continuous evolution of higher speed and more complex optical communication systems in the AI/ML era. Silicon photonics technology has become an important pathway for the "More Than Moore" era after Moore's Law, which is a clear technological development trend. Why does the bank judge that now is the time when the demand for silicon photonics modules is accelerating penetration? (1) The technological singularity has arrived. In recent years, the PDK of fabs mainly represented by Tower and Global Foundry has become increasingly mature, and the coordination between silicon photonics chip manufacturers and fab factory packaging, as well as the coordination between silicon photonics module manufacturers and downstream customers, has led to a qualitative change caused by quantitative changes, significantly improving the yield and performance of silicon photonics chips and modules. (2) Fill the EML gap. The demand for 800G optical modules will continue to increase significantly in 2025 and 2026, while upstream optical chip manufacturers mainly in the US and Japan are expanding production with a certain cycle, leading to a tight supply of 100GEML. The silicon photonics capacity bottleneck is small, which can alleviate the industry's shortage of EML. (3) Balance the relative advantages of meeting the multidimensional needs of AI/ML networks at the current stage. Silicon photonics modules are a relative optimal solution achieved after trade-offs in meeting the multidimensional demands (high bandwidth, low latency, low power consumption, stability, maintainability, scalability, flexibility, etc.) of AI/ML networks in the current stage of evolution. In the longer term, light can break through the bottleneck of high-performance computing, moving from photon integration to optoelectronic integration, spanning galaxies. Optical interconnection technology will accelerate penetration in multiple scenarios such as "device-to-device" (pluggable optical modules, CPO co-encapsulation), "board-to-board" (PCIe optical interconnect), and "chip-to-chip" (Optical I/O). (1) Silicon photonics module: evolving from traditional discrete optical modules to pluggable silicon photonics modules, reducing costs and power consumption. The bank predicts that silicon photonics modules will accelerate penetration in the 800G and 1.6T era, especially in the 1.6T era, the penetration rate of silicon photonics modules will significantly increase. (2) CPO co-encapsulation technology: By co-encapsulating EIC and PIC, the distance and latency between exchange chips and optical engines is reduced, reducing power consumption, which is an important trend in optical interconnection. The bank believes that in the Scale-out scenario (between devices), CPO is a long-term trend, but commercial large-scale use will take time. The bank predicts that CPO technology will gradually penetrate in 2026-2027 and be mass applied in the 3.2T era. (3) Optical I/O: Bringing computing chips and optical engines closer together, or directly co-encapsulating them in one package, effectively increasing bandwidth, reducing energy consumption and latency, solving the "last mile of computing" problem, and having vast future application space. Risk Warning: Development of AI large models falls short of expectations, leading to reduced demand for AI data center infrastructure construction; exacerbation of US-China trade tensions; development of silicon photonics technology falls short of expectations.