Signatures of Polaron Conduction and Trapping in the dc Performance of Thin-Film Lithium Niobate Electro-Optic Modulators

Pockels electro-optic (EO) modulators based on lithium niobate (LN) have long been the mainstay of fiber-optic communications, providing a direct connection between electronic signals and optical carriers transmitting information at the speed of light. Because of its wide transparency window (bandgap ~3.7 eV), large electro-optic coefficient (r₃₃ ~ 30.9 pm/V), and wafer-scale availability, LN is well-suited for low-loss, wide-bandwidth, and efficient modulation. Recently, the advent of thin-film lithium niobate (TFLN) on insulator, coupled with advances in dry etching, have made possible optical waveguides with much stronger optical confinement than previously achievable in bulk LN. Correspondingly, microwave electrodes can be placed closer to the optical mode without sacrificing metal-induced optical loss, improving the EO interaction strength and unlocking a new class of compact devices based on high-bandwidth and low-voltage modulation.

However, it is well-established that the EO response of TFLN modulators drifts at dc and low frequencies. Although post-processing techniques such as thermal annealing have been heuristically developed to reduce these effects, a microscopic electronic understanding of instability remains unclear. This has precluded the ability to design device improvements that target and nullify specific material origins of dc nonidealities.

Here, we present a series of elemental spectroscopy and electronic transport measurements that suggest dc EO performance can be understood from a polaronic transport and defect trapping perspective. Specifically, due to strong electrostatic interactions with the lattice, electrons in LN self-trap and transit either as free polarons hopping between reduced Nb sites or bound polarons localized to Nb_Li antisites. To this end, we show that by chemically reducing the metal-LN contact region, we can extend linearity in both the EO-voltage and current-voltage curves. This can be interpreted as changing the local polaronic conductivity, and therefore the contact resistance—reduced LN is known to exhibit semiconducting properties, compared to the insulating characteristics of unreduced LN. In contrast, SIMS measurements correlated with STEM imaging indicate that Li outdiffuses into nearby SiO₂ after annealing, leaving behind an interfacial layer of Li vacancies that can trap polarons. When implemented in a device, this observably increases the number of decay paths in an EO bias drift measurement, degrading performance.

Finally, we present preliminary evidence of a charge detrapping protocol in LN. Both I-V and EO measurements show hysteresis, where the response does not recover after repeated or extended measurement. In semiconductor electronics, this is typically interpreted as long-lived trap filling. We use this framework to implement a voltage application sequence intended to release traps, and experimentally recover both the linearity and magnitude of the EO response after 36 hours of continuous voltage application.