Identification of Thin-Film Mechanical Properties by Inverse Methods

• Main Author: 楊政達
• Other Authors: 吳恩柏
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/53938440328345693349

博士 === 國立臺灣大學 === 應用力學研究所 === 91 === In the manufacturing process of semiconductor and MEMS, the yielding is affected considerably not only by the mechanical properties but also the stress distribution of thin films. For these two factors, a series of global domain, high-resolution and real-time optical measurement system is developed in this thesis. In this system, the slope of thin film can be measured, and used to find the mechanical properties and the stress by inverse algorithm methods. This research can be separated into two parts, one is to improve the traditional reflection moiré to be Digital Phase Shifting Reflection Moiré (DPSRM), the other is to use the orthogonal slope measured by DPSRM and combine with ANSYS and the Genetic Algorithm (GA) to inversely obtain the mechanical properties. The optical calibration system includes five parts of work: (I) The slope difference of Digital Reflection Moiré (DRM) applied by Ronchi grating and sinusoidal grating is experimentally verified under 8% deviation. Thus Ronchi grating, the simplified method, is adopted in this thesis. (II) CCD is introduced to replace the traditional camera and then digitalize the traditional reflection moiré. (III) With phase shifting mechanism and phase unwrapping method, slope contour can be transformed into slope distribution, and then the whole-field, real-time DPSRM for advancing the resolution from 10-4rad/fringe to 10-6rad/pixel can be developed. (IV) By applying crossed Ronchi grating, an orthogonal slope measurement method, Crossed Digital Phase Shifting Reflection Moiré (CDPSRM) can be developed and the crossed fringe and high frequency noise can simultaneously be decoupled by FFT filters. (V) Slope distribution error can be minimized to the least by using windows of filter about 11% larer than 2 times of the central frequency. To calculate the mechanical properties of thin films, an ANSYS model and a GA optimization algorithm are developed. Unlike steepest descent method, conjugate gradient method or simplex method, the GA algorithm can eliminate the disadvantages of converging to local minimum and decrease the errors from differentiating the directional derivatives. In addition, thin film models by Airy stress function along with the GA algorithm is constructed to identify material properties of 2- or 3-layers thin film. This model is proved to be workable also for thick films. Both these two methods can detect up to four parameters of thin film properties in one sample simultaneously, such as Young’s modulus, Poisson’s ratio, coefficient of thermal expansion and thickness. In the last part, we focus on measuring the residual stress and find the magnitude of residual stress of Ni film is sputtered > electroless > electroplated. The residual stress of sputted Cu film is about -180.0MPa, and that of electroplated Cu film is about 81.6MP.