Overcoming the Bandwidth Bottleneck in Edge Data Centers via Adaptive On-Chip Optical Equalization

The explosive growth of edge intelligence and distributed computing has driven the emergence of compact data centers deployed closer to end devices, where bandwidth-hungry interconnects must operate under stringent power and cost constraints. In such environments, the prevalence of bandwidth-limited optoelectronic transceivers imposes a fundamental bottleneck on short-reach intensity modulation/direct detection (IM/DD) links, resulting in degraded signal fidelity and system scalability. Here, we demonstrate a chip-scale optical equalization approach that enables adaptive spectral reshaping in the optical domain without relying on high-power electronic processing. The device integrates thermo-optic interferometric arrays with polarization-splitting rotators, achieving polarization-insensitive, low-power, and wavelength-consistent equalization within a footprint compatible with silicon photonic transceivers. System-level evaluations with 100 Gbps PAM4 signals reveal a 50% enhancement in usable link bandwidth, up to 4 dB sensitivity improvement, and sustained performance over extended operation, while further supporting WDM-based parallel transmission at 200 Gbps. This approach offers a scalable pathway to energy-efficient edge optical interconnects, bridging the gap between device limitations and the architectural demands of future edge computing systems.