Foundry-enabled wafer-scale characterization and modeling of silicon photonic DWDM links.

AI/ML compute clusters are driving unprecedented bandwidth demands at the package boundary, motivating co-packaged integrated photonics closely co-located with the compute unit. We present a scalable silicon-photonics transceiver platform and a measurement-driven design methodology that together enable dense, energy-efficient DWDM links suitable for in-socket integration. Automated wafer-scale probing on 300 mm active photonic wafers extracts waveguide and resonator statistics using index fitting and comprehensive device characterization. The resulting wafer-scale measurements highlight design points such as wider robust waveguides, whispering gallery mode resonators, and thermally efficient undercut devices, that reduce required thermal tuning power and tighten insertion loss distributions. We propagate the measured distributions through a system model via large-scale Monte Carlo simulations to derive realistic link margins and source power targets. Together, the scalable architecture and wafer-scale measurement-informed design process offer a practical path to high-bandwidth, low energy consumption DWDM links with robust yield.