Thermo-optic phase shifter

Every interferometric photonic circuit needs phase control: to bias Mach–Zehnders, align ring resonances to channels, and trim away fabrication variation. The thermo-optic phase shifter delivers it with nothing more than a resistor: metal or doped-silicon heater above (or beside) the waveguide, current through the heater, local temperature rise, index shift $\Delta n = (dn/dT)\,\Delta T$, phase shift $\Delta\phi = 2\pi\, \Delta n_{eff} L/\lambda$.

Material contrast explains where the technique shines. Silicon's thermo-optic coefficient is large ($dn/dT \approx 1.8\times10^{-4}\ \mathrm{K}^{-1}$), so a silicon phase shifter reaches $\pi$ with modest heating: $P_\pi$ of 10–30 mW in standard geometries. Silicon nitride's coefficient is an order smaller ($\approx 2.4\times10^{-5}$), so nitride circuits pay proportionally more power for the same phase, the flip side of nitride's prized thermal stability.

Figures of merit: $P_\pi$ (power for π), response time, and optical loss (near zero, since the heater sits outside the mode; its one clean advantage over carrier-based phase shifters, whose free carriers absorb). Response is thermal-diffusion limited: microseconds to milliseconds, fine for bias and configuration, useless for data. Efficiency engineering is thermal engineering: undercut/suspended structures that block heat leakage to the substrate push $P_\pi$ below 1 mW at the cost of slower response and mechanical delicacy; folded waveguides under one heater trade length for power.

System-level realities dominate at scale. A circuit with hundreds of shifters at tens of mW each spends watts on standing still, often the largest single term in a PIC's power budget. Thermal crosstalk couples neighboring shifters (heat does not respect layout boundaries), so dense meshes need calibration matrices and feedback, with on-chip monitors closing the loop. And since ambient temperature moves the whole chip, thermo-optic bias points are servoed, not set-and-forgotten.

Alternatives when heat won't do: plasma-dispersion (fast, lossy), Pockels-effect platforms (LNOI), and MEMS or stress-optic tuners for low-power hold.