Acoustic Signals Based Classification Scheme for Assessing the Grinding Quality of SiC Wafers

Mechanical grinding is the essential processing step in manufacturing silicon carbide (SiC) wafers. To measure the quality of grinding, however, is time consuming. This paper proposes a real time measurement scheme by analyzing the audio signals collected in the wafer grinding process to determine the the grinding ratio (GR), a quality indicator. Audio signals are captured by using MEMS microphone. Fast Fourier Transform (FFT) is first applied to separate key frequency bands from others. Empirical Mode Decomposition (EMD) is next performed to extract intrinsic mode functions. Four most prominent statistical features are calculated by using Recursive Feature Elimination (RFE) [<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>] algorithm. Finally, Singular Value Decomposition (SVD) is applied to reduce the dimensionality of the feature matrix. A lightweight 1D CNN model with a minimal architecture is then developed to identify the quality of GR. The identification accuracy is up to 96%. In system implementation, an FPGA board is employed to accelerate the DSP operations and can accomplish the analysis of 1-second long audio signal (32,768 sample points) in less than 100 ms.