Ex-situ plasma diagnosis by combining scanning electron microscope, wavelet, and neural network

Abstract

Plasma processes are crucial for manufacturing integrated circuits. To maintain device yield and equipment throughput, plasma faults should be tightly monitored and diagnosed. A new ex-situ model to diagnose plasma processing equipment was presented. The model was constructed by combining wavelet, scanning electron microscope, ex-situ measurement of etching profile, and neural network. The diagnosis technique was applied to a tungsten etching process, conducted in a SF 6 helicon plasma. The wavelet was used to characterize detailed variations of plasma-etched surface. Three types of diagnosis models were constructed, trained with the vertical, horizontal, and diagonal wavelet components. For comparison, a conventional model was built by using the estimated profile data. Compared to the conventional model, the wavelet-based models, particularly the horizontal model, demonstrated a much improved diagnosis. The presented method can be effectively used to construct an improved diagnosis model for any plasma-processed surfaces.