ABSTRACT The exponential growth of artificial intelligence (AI) workloads is driving an urgent demand for optical interconnects combining ultrahigh bandwidth, energy efficiency, and scalability. Silicon photonics, with its Complementary‐Metal‐Oxide‐Semiconductor (CMOS) compatibility and wafer‐scale manufacturability, offers a promising platform for such architecture. Silicon microring modulators (MRMs), featuring compact footprints and low power consumption, are ideal for optical interconnects but remain constrained by a trade‐off between modulation efficiency and bandwidth—limiting their potential for 400 Gbps‐per‐wavelength operation. Here, we demonstrate a wafer‐level fabricated silicon MRM with a heavily doped trench‐integrated structure on a 300 mm silicon photonics platform that achieves both outstanding device performance and remarkable wafer‐scale uniformity. The device exhibits a record electro‐optic bandwidth exceeding 110 GHz (without peaking) and a practical modulation efficiency of Vπ·L ≈ 0.57 V cm. Operating in dual modes—self‐biasing for energy‐efficient scale‐up interconnects and depletion driving for ultrafast scale‐out links, the MRM supports error‐free 32 Gbps NRZ transmission over 2 km SSMF with 0.43 Vpp drive and 0 V bias, and open eyes up to 400 Gbps (PAM6). This work establishes the first wafer‐scale silicon MRM achieving 400 Gbps/λ operation, underscoring silicon photonics as a scalable and energy‐efficient platform for next‐generation AI computing networks.
Hu et al. (Fri,) studied this question.