Silicon nitride is the low-loss, CMOS-compatible dielectric that has become the default waveguide layer for integrated photonics, carrying everything from soliton microcombs driving 300 GHz wireless links to 3D optical interposers and quantum photonic meshes.
Summary
Silicon nitride (Si3N4, usually written SiN in device papers) is a dielectric thin film deposited on oxidised silicon wafers and etched into waveguides. It guides light by index contrast against a silica cladding, in the same way silicon-on-insulator does, but with three differences that decide where it is used: a wider bandgap, so it is transparent from the visible through the telecom band where silicon is opaque below roughly 1.1 µm; very low propagation loss and no two-photon absorption at telecom powers, which makes it the material of choice for high-Q resonators and nonlinear optics; and no native electro-optic effect or optical gain, so it cannot modulate or emit light on its own.
The parameters that decide a SiN design are index contrast and confinement (high confinement for compact nonlinear resonators, low confinement for ultra-low-loss delay lines and links), thickness and stress in the deposited film, thermal tuning behaviour, and how the passive layer interfaces with whichever active material provides modulation or gain. Recent work shows those parameters are still being pushed: waveguides with 50 nm cores formed by filling etched glass trenches and fusion-bonding two Borofloat wafers to give a thick symmetric cladding, two-layer 3D routing to eliminate waveguide crossings, and slab-engineered mode hybridisation in rib waveguides to cancel crosstalk between neighbours in standard foundry processes.
The honest framing is that SiN is not an emerging material. It is an incumbent substrate whose capability envelope is being extended, and whose main open question is how far active function can be added without losing the fabrication simplicity that made it attractive.
Evidence base
Open questions
- Does erbium-implanted Si3N4 reach net optical gain on chip, or does it stop at photoluminescence enhancement of roughly 18-fold with a ten-fold lifetime reduction?
- Can back-end-of-line bonding of active materials such as TFLN onto finished SiN and silicon photonics be run at commercial yield, or does the trench-based die-to-wafer approach stay a laboratory route?
- Once dispersive thermo-refractive coefficients are included, with 7% variation measured across a telecom-to-visible octave, how much do existing octave-spanning microcomb and phase-matching designs need to be revised?
- Does the glass fusion-bonding route for 50 nm-core low-confinement waveguides survive the transition from straight test structures to functional circuits with bends and couplers?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.