Manufacturing Technology Today

Ranga Komanduri ¹, Prahalad Rao ², Prakash Vinod ³, Omer Beyca ², Sanjay Jagannath ⁴, Adam Fields ⁵, Zhenyu Kong ⁵, Satish Bukkapatnam ⁵

Affiliations
1 School of Mech., & Aerospace Engg, Oklahoma State University, Stillwater, Oklahoma, US
2 School of Industrial Engg., & Management, Oklahoma State University, Stillwater, Oklahoma, US
3 Central Manufacturing Technology Institute, Bangalore, IN
4 Dept. of Mech. Engg., Indian Institute of Technology, Chennai, Tamil Nadu, IN
5 School of Industrial Engg., & Management, Oklahoma State University, Stillwater, Oklahoma, IN

Abstract

This document presents the battery of tests conducted to ascertain the physical sources of dominant vibration sensor signal patterns observed in chemical mechanical planarization (CMP) process. A Buehler (model Automet® 250) bench top CMP machine is instrumented with miniature MEMS 3-axis accelerometer (Analog Devices ADXL 335) and audio sensors (Analog Devices ADMP 401). The CMP setup is used for finishing blanket copper workpieces to a surface finish of Ra ~ 5 nm. While the sensor signals are sensitive to variations in the CMP process, the extraneous noise prevents the direct use of raw signal patterns for early detection of defects. Consequently, instead of primarily monitoring the raw sensor signal patters, we isolate signal features which are indicative of process state from those which are mere artifacts, and thus potentially valuable for process monitoring.

The frequency spectrum of typical MEMS vibration sensor signals acquired during the CMP process contains three dominant components, namely:

1. Low frequency component in the 0.5 - 1 Hz region.

2. Broadband frequency regions centered around 25 Hz and 50 Hz.

3. Broadband frequency region around 120 Hz.

A total of nine tests are reported to ascertain the underlying physical cause of each of these components. Component 1 is shown to result from eccentricity errors in the polishing head (spindle); component 2 is most likely a conjoined effect due to sensor characteristics, electromagnetic interference from machine elements, and structural vibration; component 3 is observed to respond to changing downforce (polishing load) conditions, and process state, such as pad wear, and as such is useful for process monitoring applications. In addition, a relatively small (-110 dB) background (white) noise is evident throughout the frequency spectrum of CMP vibration signals; this is explained as originating from measurement errors and environmental factors.

Full Text

     

Satish Bukkapatnam ¹

Affiliations
1 Oklahoma State University, Stillwater, OK, USA, US

Abstract

Much of the complexity in grinding and other abrasive finishing processes emerges from nonlinear stochastic dynamics of the underlying processes. Earlier attempts to consistently monitor surface quality using conventional, largely statistical and linear dynamic, approaches have had limited success. In this paper, the analytical derivation and experimental validation of an approach for real-time monitoring of surface finish (R_a < 50nm) in precision grinding are presented. The approach is based on reconstructing customized multi-scale representations (CMR) and low-order nonlinear stochastic differential equation (n-SDE) models from an array of vibration sensors. The experiments were conducted on an instrumented surface grinder (Proth PSGS 3060BH) and a cylindrical grinder (Browne and Sharpe 1024). Vibration sensor features extracted based on the models were used to estimate R_a under various wheel degradation levels and in-feed settings. The results show that model correctly captures the dimensionality of the dynamics underlying the measured vibration signals. The features were able to accurately estimate R_a values below 50 nm, and improve surface roughness predictability by∼30% over commonly used static mappings between the features and quality states.

Keywords

Dynamic, Nano, Monitoring.

Full Text