Development of AlN based Piezo-MEMS Technology: Film Growth to Device and Application

Abstract

Aluminum nitride (AlN) has emerged as an excellent material for a variety of MEMS applications owing to its unique combination of properties such as- low dielectric constant, strong polarity, low dielectric and acoustic losses, high acoustic velocity, excellent stability and CMOS compatibility. This thesis presents the development of AlN based piezo-MEMS technology, emphasizing film growth, material characterization, device design, fabrication and device characterization. The first part of the work focuses on developing a two-step growth technique to achieve thick, high-quality AlN films on a Si (111) substrate for piezo-MEMS applications. In this approach, a high-quality AlN seed layer is first deposited using metalorganic chemical vapor deposition (MOCVD), followed by the growth of a thicker AlN layer by sputtering. This method effectively overcomes the limitations of each individual technique- specifically, the thickness constraints of MOCVD and the inferior crystal quality typically associated with sputtered films. Material, electrical and mechanical characterizations demonstrate the superior quality of the two-step AlN film compared to films grown solely by sputtering. The second part of this work presents a novel cantilever geometry for AlN based sensing that incorporates a stub extension at the tip mass. This design enables a reduction in resonant frequency without increasing the device footprint, while simultaneously enhancing piezoelectric charge generation. To further optimize the cantilever design, an AI/ML based design optimization methodology is used, targeting applications such as a bio-mimetic MEMS cochlea and a real-time Fourier-like transform. A Gaussian Process Regression (GPR) model, trained on finite element method (FEM) simulation data, is employed to accurately predict device performance. This model is integrated with optimization algorithms, namely fmincon and genetic algorithms, to identify geometric parameters of the device that achieve the target resonant frequency while maximizing charge output within a constrained area. The resulting optimized designs demonstrate output voltages in the range of 2-2.4 V sufficient for direct neural stimulation. The third part of this work focuses on the development of a multi-device, multiuser fabrication platform capable of realizing multiple AlN-based MEMS devices on a single Si or SOI (111) wafer. During fabrication, several challenges were encountered— most notably, the high tensile stress in the AlN films, which led to the failure of all devices in early fabrication runs. The process modifications implemented to address these issues are discussed. Additionally, initial characterization results are presented for devices fabricated using this platform- a cantilever array designed for Fourier-like transform implementation, and a high-bandwidth accelerometer. Finally, a new technique for direct measurement of spontaneous polarization (𝑃𝑠) in polar thin films is proposed and experimentally validated. This method estimates spontaneous polarization from the resonant and super-harmonic resonant strain response of a thin-film piezoelectric-on-silicon (TPoS) resonator, while also accounting for the dielectric non-linearity of the material. This method enables the direct measurement of 𝑃𝑠 in non-ferroelectric thin films for the first time. The technique was first validated using lead zirconate titanate (PZT) by benchmarking the extracted spontaneous polarization value against that obtained from standard polarization–electric field (P–E) loop measurements. It was then applied to aluminum nitride (AlN), a non-ferroelectric material at room temperature, yielding a spontaneous polarization value of 0.18–0.2 C/m².