Uncovering the formation history of dusty star-forming galaxies in the early universe requires wideband spectroscopic instruments capable of detecting redshifted emission lines in the millimeter and submillimeter regime. The on-chip spectrometer DESHIMA 2.0 addresses this by integrating filterbanks and kinetic inductance detectors (KIDs) onto a single chip, enabling wide frequency coverage in a compact footprint. However, performance is currently limited by losses in the dielectric layer of its microstrip lines, which reduce transmission efficiency and sensitivity. Minimizing these losses is crucial not only to enhance signal throughput but also to enable the use of parallel-plate capacitors (PPCs), which significantly reduce resonator size, increase detector count, and thus improve spectral resolution. Achieving low-loss dielectric films is therefore essential for the next generation of high-resolution spectrometers such as TIFUUN.

To support this development, this thesis focuses on optimizing the deposition of hydrogenated amorphous silicon (a-Si:H) films using Inductively Coupled Plasma Enhanced Chemical Vapor Deposition (ICP-CVD). These films function as dielectric layers in both microstrip lines forming the filterbank (90–360 GHz) and in PPCs (1–10 GHz).

Since direct measurement of dielectric losses was beyond the scope of this project, optimization was based on properties known to correlate with loss mechanisms—namely two-level systems (TLSs) and absorption in the vibrational tail. TLSs are expected to dominate at PPC operating frequencies, while vibrational absorption is more relevant in the filterbank. The optimized properties include residual stress, thickness non-uniformity, optical and infrared refractive index, band gap, void-volume fraction, hydrogen content, and microstructure parameter.

A Taguchi L18 orthogonal array was used to systematically vary seven deposition parameters: table temperature, silane flow rate, ICP power, table RF power, gas ratio (silane/argon), pressure, and native oxide removal method (argon milling vs HF dip). Film properties were evaluated using ellipsometry, FTIR spectroscopy, and a stressmeter.

ANOVA revealed table RF power as the dominant factor, with the highest contribution to five of the eight properties: thickness uniformity, refractive index, hydrogen content, microstructure, and residual stress. In the latter, increasing power shifted the film from tensile to strongly compressive regimes. Other parameters had more moderate effects: chamber pressure had strongest influence on void-volume fraction and infrared index, while silane flow rate and wafer preparation affected the band gap most significantly. Despite some models exhibiting high residuals, indicating unmodeled interactions between the parameters, the overall analysis successfully identified key relationships between the deposition parameters and the material properties.

Two optimized recipes were selected using Grey Relational Analysis, with equal weighting assigneds to hydrogen content and void-related properties. Recipe 19 achieved minimal hydrogen content (4.3 at.%) and low residual stress, making it a strong candidate for minimizing dielectric losses in the filterbank. However, it also showed high void fraction and surface inhomogeneity. Recipe 20, aimed at maximizing hydrogen content, reached 16.3 at.% but showed only average performance and suffered from high compressive stress (–745 MPa). As a more robust alternative, Recipe 9 offered low void content, moderate hydrogen level, favorable stress (+125 MPa), and good uniformity—making it the most practical candidate for integration into both filterbank and PPC structures.

Future work should focus on direct measurement of dielectric loss tangent and TLS density under cryogenic conditions to validate the predicted performance in superconducting spectrometers.