Thin-Film Lithium Niobate-based Photonic Integrated Circuits - Development of Fabrication Processes and Modulators

Abstract

Lithium Niobate (LN) has many advantages such as low optical loss, a large electro-optic (EO) coefficient, and a large nonlinear optical coefficient. Recently, thin-film LN (TFLN)-based photonic integrated circuits (PICs) have been mainly investigated because the structure of TFLN on insulator enables small-area waveguides with a strongly confined mode, which lead to high-performance, large- scale PICs. My thesis focuses on the development of fabrication processes for TFLN-based optical modulators, including the establishment of fabrication methods, the design and characterization of passive photonic components and Mach–Zehnder modulators (MZMs) based on a traveling wave electrode and a lumped electrode configuration.

Chapter 1 introduces needs for PICs, material properties of LN, various LN waveguide structures, and research motivation of this thesis.

Chapter 2 describes the complete fabrication process flow and provides detailed explanations of individual steps. To eliminate sub-field stitching errors in e-beam lithography (EBL), the values of subfield trim, dose, rotation and shear parameters were optimized. Also, a bilayer process using lift-off resist (LOR) was employed to form high-quality chromium (Cr) etch masks for waveguide patterning. Before deposition of Cr, I found that oxygen (O2) plasma treatment is effective to get a high yield of the Cr lift-off process. The effects of ICP and RIE power as well as chamber pressure on argon (Ar)- based LN dry etching were investigated. During the development of the dry etching technique, aluminum (Al) contamination was identified as a major issue, and its origin was investigated. By modifying etch conditions including ICP power, RIE power, chamber pressure and introducing chlorine (Cl2) gas which is effective for removing Al, the contamination was reduced to a negligible level. After that, various etching tests confirmed that the etching selectivity between LN and Cr is approximately 6:1, indicating that an 80-nm-thick Cr mask is sufficient for etching 300 nm of LN. A scanning electron microscope (SEM) image showed severe sidewall roughness caused by redeposition of LN particles during the dry etching. To address this issue, cleaning using an RCA-1 solution was introduced, which effectively smoothed the sidewalls without causing a decrease in the LN thickness when the ratio of DI water, ammonium hydroxide (NH₄OH), and hydrogen peroxide (H₂O₂) was carefully optimized. Finally, due to LN crystal damage that could occurring in TFLN wafers, EO modulation efficiency is very low without annealing treatment. I confirmed that thermal annealing is required to restore EO modulation efficiency from the experiment, comparing the measured half-wave voltages of annealed and non- annealed traveling wave Mach-Zehnder modulator (TWMZM) samples.

Chapter 3 explains on the design and experimental results of passive photonic devices, including grating couplers (GCs), multimode-interference (MMI)–based 1×2 power splitters, MMI-based waveguide crossings (WGCrses), and 90° Euler curve-based bent waveguide (ECBW). These components were designed as basic building blocks for meandered waveguide and lumped element- based MZM (MLMZM), introduced in chapter 4. Metallic focusing grating couplers (MFGCs) were developed to easily fabricate grating couplers on TFLN and reduce the area of one-dimensional (1D) metal grating couplers (MGCs), and titanium dioxide (TiO₂)-assisted MFGCs (TMFGCs) were also developed to enhance coupling efficiency (CE), by increasing constructive interference of light in waveguide core using 80 nm-thick TiO2. Experimental results show CEs of −6.23 ± 0.31 dB for MFGCs and −5.25 ± 0.23 dB for TMFGCs, demonstrating a 0.98 dB improvement in the average CE, enabled by the TiO₂ additional layer. The MMI-based 1×2 power splitter was designed using the self-imaging theory, and 50:50 splitting performance was verified experimentally after device-length optimization. For the design of the MMI-based WGCrs, the anisotropic refractive indices of LN were precisely considered, depending on Y- and Z-propagating directions. The simulated and measured insertion losses (ILs) in the case of Y-propagation are 0.0355 dB/cross and 0.034 dB/cross, respectively; those in the case of Z-propagation are 0.0584 dB/cross and 0.08 dB/cross, respectively. A good agreement between the simulation and experiment was achieved. In the design of 90° ECBWs, the bent waveguide exhibited an IL below 0.2 dB/90° for an effective radius of 50 μm, and measurements confirmed a loss of 0.196 dB/90°.

Chapter 4 provides the theoretical background of the EO phase modulation in TFLN, including the EO coefficient tensor and the equation for an effective index change, derived from perturbed Maxwell’s equations. The calculation of the VπL product and the dependence of the product on a waveguide width and an electrode spacing were explained. The push-pull operation of MZMs was explained, followed by the design, measurement, and analysis of TWMZMs. To understand the RF characteristics of TWMZMs, the effects of various geometrical parameters, such as the buried oxide (BOX) thickness and the signal electrode width and thickness, on the characteristic impedance, effective index, and RF attenuation of the traveling wave electrode were investigated. From DC and low-frequency AC measurements, VπL values of 1.09 V·cm and 1.19 V·cm were extracted, respectively. The predicted 3- dB EO bandwidth of 10.51 GHz was in good agreement with the experimentally measured value of 10.38 GHz. To realize more compact modulators as compared to TWMZMs, the MLMZMs were designed and experimentally demonstrated. Two versions of MLMZMs were developed. The first version employed straight WGCrses, single 50 μm-radius 90° circular bends, and separated top and bottom electrodes enabling the push-pull operation. Measured VπL values were 1.596 V·cm for a 2.1-mm-long device and 1.476 V·cm for a 3.6-mm-long device, with both devices exhibiting 3-dB EO bandwidths of only 1.18 GHz. The simulated capacitance of the 1st version structure was approximately 2.5 pF. This large capacitance explains the limited bandwidth. The second version incorporated the improved passive components such as the MMI-based WGCrses and the 90° ECBWs, introduced in chapter 3, along with modified electrode geometries to reduce the total device capacitance. Equivalent-circuit modeling predicted bandwidths up to 8.65 GHz for a 1.5-mm device and 3.78 GHz for a 3.35-mm device. Experimental measurements confirmed 6.82 ± 0.87 GHz and 2.36 ± 0.63 GHz for the 1.5-mm and 3.35- mm devices, respectively. From this experiment, I confirmed that the order of picosecond response time is achievable. Footprint of devices within 1 mm2 is achievable using the design of MLMZM. Additional analysis was performed to evaluate the influence of source impedance and parasitic resistance on the EO bandwidth. This thesis builds a comprehensive technological foundation for compact, high-speed TFLN modulators through developments in fabrication techniques, passive photonic components, and modulator design strategies. The presented results offer practical design insights for further scaling and performance enhancement of TFLN photonic platforms. Moving forward, promising research directions include monolithic integration of RF drivers to suppress parasitic effects, utilization of inductive-peaking techniques, exploration of advanced high-speed electrode configurations, and fabrication processes that yield smoother, lower-loss waveguides. Collectively, these efforts will accelerate the realization of next-generation ultra-compact photonic processors and high-capacity communication systems based on TFLN technology.