Tower Semiconductor announced on November 3, 2026 that Innolight’s first near-package optics (NPO) device has entered early production on Tower’s PH18M silicon photonics process in California, while multiple NPO products for future generations have taped out on its 300mm platform.
The work combines Innolight’s optical interconnect development with Tower’s high-volume silicon photonics manufacturing to address rising bandwidth and performance demands in AI infrastructure. Innolight chief technology officer Oliver Sun said the collaboration is intended to accelerate NPO development for hyperscale and AI data-center customers.
Tower president Marco Racanelli described Innolight as a cornerstone manufacturing customer across four Tower facilities: Newport Beach, California; San Antonio, Texas; and one site each in Israel and Japan.
Higher Channel Counts Raise Manufacturing Requirements
Tower said NPO architectures typically require substantially more optical channels than conventional pluggable modules, often integrating 16 or more channels on one device, and could add demand beyond today’s pluggable market as AI workloads scale. As channel counts rise, manufacturing expertise, process control and yield performance become increasingly important.
The platforms support monolithic integration of multiple photonic functions, including wavelength-division multiplexing and demultiplexing, on a single chip. That integration allows more functionality within a compact device, placing greater weight on process uniformity and yield as designs become more complex.
Tower’s silicon photonics platform documentation lists low-loss silicon and silicon-nitride waveguides, germanium PIN diodes and avalanche photodiodes, phase shifters, Mach-Zehnder modulators and on-chip heaters, with options for on-chip or integrated lasers. It also includes foundry design kits, multi-project-wafer prototyping and in-line wafer-level optical testing. Together, those features supply the component and process base for the denser photonic designs Tower described.
On Tower’s August 4, 2026 earnings-call transcript, chief executive Russell Ellwanger said NPO delivers greater bandwidth density and lower energy per bit than pluggable optics while using the same established ecosystem to address operator concerns about reliability, serviceability and multisource flexibility.
An Established Silicon Photonics Partnership
The NPO milestone extends an existing manufacturing relationship. The companies’ September 7, 2023 partnership announcement covered multi-generation transceivers based on Tower’s PH18 process, with production already underway. Innolight had launched multiple 400G and 800G products on PH18M; 400G parts were in mass production, while 800G volume production was scheduled for the fourth quarter of 2023.
Innolight’s AI and data-center networking page identifies high bandwidth, low power and low latency as cloud-operator requirements as generative AI accelerates optical-network bandwidth growth. NPO extends that optical roadmap into devices integrating more channels and functions on one chip.
On the same August call, Ellwanger said several NPO deployments were planned over the following year and more were in design. His forecast for 2027 was that NPO would become significant and reach a shipment share in the tens of percent, especially in the second half, although the exact percentage was not yet known.
The central question is whether Tower can sustain the process control and yield required as higher-channel-count devices move from early production into later generations. That manufacturing execution will determine how the companies’ expanding NPO design work aligns with the deployment and shipment expectations management has set.