Global Tech News For people figuring out what to build with AI.
Chips

Transferred Single-Crystal Membranes Add Functions to Silicon Photonics

Close-up of an indium phosphide photonics multi-project wafer

Researchers have demonstrated separate light modulators, isolators and detectors by transferring thin single-crystal membranes onto silicon and silicon-nitride photonic chips. Moving the materials after they are grown avoids the atomic-lattice mismatch that can make it difficult to grow one crystal directly on another.

Silicon (Si) and silicon nitride (SiN) route light through compact waveguides, but they do not provide every optical function. The team grew each functional crystal separately, freed it as a thin membrane, then transferred it onto a patterned Si or SiN circuit. This lets the crystal keep its structure while adding its optical behavior to an established platform.

For one demonstration, barium titanate (BTO) served as the electrically active material in a silicon light modulator. Applying voltage changes BTO’s refractive index and shifts the phase of light. A Mach–Zehnder modulator uses that phase shift to change the interference as two light paths recombine. From low-frequency device measurements and optical-field modelling, the authors estimate BTO’s Pockels coefficient—a measure of how strongly voltage changes its optical properties—above 1,290 picometres per volt. A separately designed high-speed modulator measured an electro-optic 3-decibel bandwidth above 23 gigahertz: its response had not fallen to half its reference power at 23 gigahertz. The two results came from different layouts; the study does not show one modulator combining both results. This bandwidth measurement is not a demonstrated data rate.

Other prototypes used cobalt ferrite (CFO) membranes in silicon microring resonators to suppress light traveling in the reverse direction, a function called optical isolation. Gallium arsenide (GaAs) and gallium nitride (GaN) membranes were placed on silicon-nitride photonics for photodetection spanning ultraviolet to near-infrared wavelengths. These were distinct device demonstrations, not three functions operating together on one chip.

The study is a laboratory proof of concept, not evidence of foundry production or manufacturing yield. The supplementary information says imperfect fabrication of the electrodes carrying high-speed signals limited the modulator; it estimates 6 to 8 decibels of light-coupling loss at each chip facet and reports total fiber-to-fiber loss of 18 to 22 decibels. Production yield was not demonstrated.

Sources