RF-over-dielectric-waveguide interconnect

Cross-cuts: Memory
last updated 2026-07-16
Optical InterconnectOptical InterconnectHBM (High-Bandwidth Memory)HBM (High-Bandwidth…RF-over-d…

Data transmission by modulating a millimetre-wave / sub-THz RF carrier onto a plastic dielectric waveguide instead of driving copper conductors or optical fibre. The transmitter and receiver are CMOS/SiGe RF SoCs; the channel is a low-loss dielectric “tube” or ribbon. The pitch: copper’s reach collapses above 200G/lane while optics carries laser, photonic assembly, and reliability cost; a purely electrical RF front-end with a waveguide channel claims the mid-range (roughly 1-10m, in-rack and adjacent-rack) socket at optics-class bandwidth without optics-class complexity.

Two funded proponents as of Q2 2026 (see ref):

Why this page exists. The medium is a live counter-datapoint to the load-bearing assumption Optical Displaces Copper: it attacks the same reach class that co-packaged and near-packaged optics are supposed to take next, with a technology that keeps the electrical ecosystem (SerDes, retimers, cable vendors) intact. It does not need to beat optics everywhere to matter; it only needs the mid-range socket to delay the optical crossover by a reach class. Tracked so the radar and cross-pollination can tag papers, funding, and standards activity in this lane. Evidence lands on the assumption’s evidence_log and on Scale Up Interconnect.

Recent mentions

Related concepts

Frontier questions