Towards Scalable Electro-Optic Systems in Thin Film Lithium Niobate Integrated Photonics

Abstract

Thin-film lithium niobate integrated photonics is a compelling platform for highly integrated electro-optic systems. The platform features compact, high-bandwidth phase modulators which can operate at CMOS compatible voltages, together with low optical loss, enabling photonic systems with large numbers of cascaded modulating elements. For instance, thin-film lithium niobate is an excellent candidate for optical switch networks (photonic circuits composed of large arrays of voltage-controlled optical phase shifters), which are at the core of powerful emerging photonic technologies, including high-performance optical computers, optical quantum computers, and datacenter interconnects.

Lithium niobate electro-optic modulators are celebrated for their high-frequency performance (f > 1 GHz), but historically have struggled with stability at the lowest frequencies (f 1 Hz). For example, lithium niobate modulators can suffer from electro-optic bias drift, where the phase of the optical carrier drifts under the application of a DC voltage. The mitigation of electro-optic bias drift in thin-film lithium niobate is a challenging problem with enormous technological importance. Bias-drift is undesirable for highly scaled switch networks, which rely on the simultaneous stable tuning of thousands of optical interferrometers. Thermo-optic tuners have been introduced as a possible workaround, but with a static energy consumption cost per modulator of order 10 milliwatts (a factor 10^6 larger than electro-optic biasing), a penalty that translates to kilowatts in highly scaled networks. In bulk LN modulators, electro-optic drift has been addressed. Decades of research have led to drift mitigation techniques which permit the stable operation of modulators with an electrical bias over long timescales.

Drift-mitigation in lithium niobate modulators is a challenging problem in optical material engineering and device physics. There are many possible drift mechanisms, overlapping in frequency response, each sensitive to a huge set of device parameters (material properties, device geometry) that may vary by orders of magnitude during modulator fabrication. Currently, we are in the early stages of understanding the low-frequency response of the thin-film modulator platform.

In this work, we present a systematic exploration of the low-frequency response of etched thin-film lithium niobate modulators. Building on drift-mitigation work in bulk modulators, we explore the effect of drift-mitigation techniques on the bias-drift observed in the thin-film platform. Moreover, we identify and address new low-frequency behavior specific to thin-film electro-optic modulators.

One problem unique to the thin-film modulator platform has been a lack of DC response altogether in some devices. We illustrate this effect by inserting an oxide buffer layer between lithium niobate and the metal electrodes. We study the low-frequency response of this system down to cryogenic temperatures, and introduce a preliminary dielectric relaxation model to explain the response. We posit a model which relies on enhanced lithium niobate surface conductivity to explain the strong suppression of the DC response of optical modulators to electrical bias. The DC performance can be restored by removing the oxide buffer layer.

Further, we illustrate the sensitivity of the low frequency response to the contact condition between lithium niobate and the metal electrodes, and show how a careful combination of etch chemistry and etch depth can improve DC performance.

Finally, we observe a distinct lack of frequency flatness in the sub-MHz response of TFLN modulators, including a broad enhancement in response around 1 kHz. We find that this feature is insensitive to the electrode interface condition. We posit that these features be photorefractive in nature, but find that doping of the lithium niobate with magnesium oxide (a technique which suppresses the generation of photorefractive carriers) does not improve the frequency flattness. We find that high temperature annealing of the lithium niobate in oxidizing or reducing environments can improve the frequency flattness across the 1 mHz to 1 MHz frequency range.

In support of this work, we present high-resolution imaging and material analysis of etched lithium niobate waveguides. We report on the interface roughness of the material after etching. This work provides insight into the crystal quality and defect structure of the material in support of low-frequency studies.

This work paves the way towards highly integrated electro-optic systems, including important demonstrations of improved low-frequency performance. Future work to model and refine these techniques, and implement them repeatably at wafer-scale will open the door to large scale switch networks in thin-film lithium niobate.