Wafer-scale waveguide sidewall roughness scattering loss characterization by image processing

TL;DR

This paper presents a scalable image processing method using SEM images to characterize waveguide sidewall roughness and estimate optical scattering loss, achieving high accuracy and a record low loss at visible wavelengths.

Contribution

It introduces a novel 2D SEM image-based approach for wafer-scale sidewall roughness characterization, replacing costly AFM measurements.

Findings

Achieved a record low loss of 0.075 dB/cm at 633 nm

100% success rate in edge detection for roughness estimation

Validated the method against theoretical Payne model

Abstract

Photonic integrated circuits (PICs) are vital for developing affordable, high-performance optoelectronic devices that can be manufactured at an industrial scale, driving innovation and efficiency in various applications. Optical loss of modes in thin film waveguides and devices is a critical measure of their performance. Thin films growth, lithography, masking, and etching processes are imperfect processes that introduce significant sidewall and top-surface roughness and cause dominating optical losses in waveguides and photonic structures. These roughness as perturbations couple light from guided to far-field radiation modes, leading to scattering losses that can be estimated from theoretical models. Typically, with UV-based lithography sidewall roughness is found to be significantly larger than wafer-top surface roughness. Atomic force microscopy (AFM) imaging measurement gives 3D and high-resolution roughness profile but the measurement is inconvenient, costly, and unscalable for large-scale PICs and at wafer-scale. Here, we evaluate the sidewall roughness profile based on 2D high-resolution scanning electron microscope imaging. We characterized the loss on two homemade nitride and oxide films on 3-inch silicon wafers with 12 waveguide devices on each and co-related the scattering loss estimated from a 2D image-based sidewall profile and theoretical Payne model. The lowest loss of guided fundamental transverse electric (TE$_{0}$) is found at 0.075 dB/cm at 633 nm across 24 devices, which is a record at visible wavelength. Our work shows a 100% success in edge detection in image processing to estimate autocorrelation function and optical mode loss. These demonstrations offer valuable insights into waveguide sidewall roughness and comparison of experimental and 2D SEM image-processing-based loss estimations.