SILITH and UMC move 1.6T silicon photonics to mass production

Mass production of 1.6T silicon photonics signals a shift from architectural theory to hardware reality for next-generation data center fabrics. By utilizing UMC's existing 12-inch CMOS-compatible foundry capacity, SILITH circumvents the scalability bottlenecks common in specialized manufacturing, offering a lower-cost path to 1.6T throughput. For the streaming ecosystem, this transition is critical for high-density AI training and real-time video processing pipelines that are quickly outrunning the thermal and bandwidth limits of traditional copper-based interconnects. Watch for initial volume deployment data from UMC's first tier-one cloud partner in late 2024 to gauge real-world yield performance.

The push toward high-volume silicon photonics arrives as the industry grapples with the 'memory wall' and high energy consumption in AI clusters. Per ByteIota (December 2025), Marvell recently committed up to $5.5 billion to acquire Celestial AI, a move intended to secure 'Photonic Fabric' technology that replaces electrical interconnects with light to connect GPUs and memory. This reflects an intensifying arms race between Marvell, Broadcom, and Nvidia to integrate optical I/O directly into compute packages. Marvell also demonstrated a 1.6T optical DSP with 200 Gbps interfaces in March 2024 at the OFC conference, underscoring that 1.6T has become the new benchmark for hyperscale networking.

Market analysis from Mordor Intelligence (June 2026) suggests the silicon photonics sector will reach $13.18 billion by 2031, with 300mm wafer category accounting for over 62% of market value. This growth is driven by the fact that optical links consume significantly less power than copper—roughly 6.2 picojoules per bit compared to over 60 picojoules for traditional electrical connections like NVLink, according to ByteIota reporting. This energy efficiency is pivotal as global data center power consumption hit 460 TWh in 2024, prompting tighter regulatory oversight and a surge in demand for low-power optical interconnects.

Beyond current silicon-based modulators, UMC is already exploring thin-film lithium niobate (TFLN) solutions for future ultra-high-bandwidth needs. This aligns with broader industry movements where competitors are eyeing co-packaged optics (CPO) to reduce the physical distance data must travel. Per Substack-based BEP Research (March 2026), startups like Ayar Labs have secured backing from Nvidia and Intel to develop chiplets that move optical interfaces onto the processor package itself. UMC's entry into the 12-inch foundry space for photonics adds much-needed merchant capacity as the supply chain shifts from pluggable modules toward these highly integrated architectures.