Key Takeaways:
The first mass-produced silicon photonics wafers have rolled off UMC's Singapore 12-inch fab, marking a shift from development to high-volume manufacturing for a technology that lets chips communicate at speeds needed for AI data centers. SILITH Technology, a Singapore-based fabless company, and UMC delivered the photonic integrated circuits after 18 months of joint development, with the platform already qualified by a leading cloud infrastructure customer for volume deployment.
"This closes the gap between silicon photonics innovation and the manufacturing scale that AI infrastructure demands," Jason Zhang, chief technology officer at SILITH, said. "Together with UMC, we are bringing together leading-edge silicon photonics design and high-volume 12-inch manufacturing to deliver the performance, scalability and cost efficiency required for the next generation of AI networks."
The platform supports 1.6T optical interconnects using 200G/lane technology, with cumulative shipments of more than 8 million 100G/lane and 200G/lane PICs already in the field. SILITH and UMC are now extending the roadmap to 400G/lane using pure-silicon Mach-Zehnder modulators, preserving the manufacturability and cost advantages of CMOS-compatible silicon. UMC is also making its own 12-inch silicon photonics platform available for customer product development in 2027, and is collaborating with ecosystem partners on thin-film lithium niobate solutions for future ultra-high-bandwidth optical interconnects.
The milestone matters because silicon photonics is becoming a foundational technology for AI data center infrastructure, where the bottleneck has shifted from compute to connectivity. Optical interconnects based on silicon photonics can move data at terabit speeds while consuming less power than traditional copper-based links, making them critical for scaling AI clusters. UMC, with 12 fabs and combined capacity of more than 400,000 wafers per month, is positioning its Singapore facility as a key manufacturing hub for this emerging market, competing with TSMC's advanced packaging capabilities in the optical interconnect space.
Why Silicon Photonics Matters for AI Infrastructure
Traditional copper interconnects hit physical limits at higher data rates — signal degradation, heat, and power consumption all increase as speeds rise. Silicon photonics solves this by using light instead of electricity to move data between chips, enabling 1.6T and eventually 3.2T links that AI clusters need to keep thousands of GPUs working in parallel. The technology also supports co-packaged optics, where optical engines sit directly next to switch ASICs, reducing the power and latency of pluggable transceivers.
SILITH's approach uses standard CMOS-compatible silicon, meaning the photonic ICs can be manufactured on existing 12-inch fab lines without specialized equipment. This is a cost advantage over alternatives such as indium phosphide or thin-film lithium niobate, which require dedicated processes. UMC's SOI (silicon-on-insulator) manufacturing capabilities, proven across its 12-inch fabs, provide the yield and reliability that cloud customers require for volume deployment.
Competitive Landscape and Investment Angle
The silicon photonics market is attracting major foundry investment as AI data center demand accelerates. TSMC has been developing its own silicon photonics platform through its 3DFabric advanced packaging roadmap, while Intel's integrated photonics group has shipped silicon photonics products for years. UMC's entry with a qualified, mass-production platform gives cloud customers an alternative supply source, potentially reducing single-source risk in the optical interconnect supply chain.
For UMC, the silicon photonics ramp represents a new revenue stream beyond its core mature-node foundry business, which faces pricing pressure from capacity oversupply. The company trades on the NYSE under UMC and on the Taiwan Stock Exchange under 2303. SILITH, founded in 2021, has secured design wins with leading cloud infrastructure and optical networking customers, validating its design-to-manufacturing model. The 400G/lane platform, if successful, could extend the company's competitive advantage against peers developing multi-chip photonic solutions using more exotic materials.
UMC is also exploring thin-film lithium niobate modulators for future 400G/lane and beyond applications, which would combine the electro-optic efficiency of lithium niobate with the scalability of silicon manufacturing. Combined with UMC's advanced packaging technologies, these platforms could enable optical-engine modules for co-packaged optics and optical I/O architectures in next-generation AI infrastructure.
This article is for informational purposes only and does not constitute investment advice.
