Machine Learning-Driven High-Throughput Analysis of Damping Effects in Silicon-Based Cantilever Resonators with Metallic Coating

Abstract

Micromechanical and nanomechanical resonators serve as essential components in a wide range of applications, including resonant sensing and energy harvesting. Resonance testing plays a fundamental role in characterizing the dynamic behavior of MEMS devices, providing insights into critical properties such as energy dissipation and structural integrity. Automated high-throughput resonance testing of silicon-based cantilever resonators with various metallic coatings offers valuable insights into the damping effects on resonance frequency and quality factor. A comprehensive study is conducted on a full wafer containing 580 devices, each hosting four cantilevers with varying dimensions, to evaluate damping mechanisms. These cantilevers are analyzed in both uncoated (reference) state and with metallic coatings of varying coverage percentages and materials, including aluminium, platinum, and molybdenum. Machine learning (ML) serves as an effective tool for interpreting complex patterns in large datasets, facilitating the identification of trends and dominant damping mechanisms that are otherwise difficult to discern. A ML framework is applied to analyze the extensive dataset, incorporating features such as cantilever dimensions, die location on the wafer, and coating parameters. Predictive models are developed to estimate changes in resonance frequency and quality factor while isolating the dominant damping mechanisms. Particular attention is given to the contributions of thermoelastic damping and surface effects, which have a significant impact on the dynamic performance of silicon cantilevers. This research highlights the importance of integrating automated high-throughput testing with ML to enhance the understanding of damping mechanisms in MEMS resonators. The results provide a foundation for the design of high-performance cantilever resonators tailored to specific applications, particularly those requiring the minimization of damping losses.