Extremely vertical sidewall trench etching on silicon substrate and modelling etching using artificial neural network

Abstract

The control of etching parameters e.g. RF power, gas flow rate, chamber pressure and bias voltage is critical for the modern semiconductor fabrication industry. It is quite benefit to clarify the correlation between parameters and predict the etching results on both scientific research and industrial fabrication. In this paper, we performed a combination of experimental and machine learning analysis to understand the importance of each parameter and the correlation between them. SF6 and O2 was employed in order to investigate the optimal condition for Si trench etching with vertical sidewall based on capacitively coupled plasma reactive ion etching system (CCP-RIE). Scanning electron microscopy (SEM) and atom force microscope (AFM) were separately applied to characterize the etching profile and measure the etching depth. According to the experimental results, the optimal recipe of etching trench with vertical sidewall on silicon substrate is proposed to be a mixture of 30 sccm SF6 flow rate and 20 sccm O2 flow rate with 240 W RF power under the pressure of 5.6 Pa. Machine learning method of artificial neural network (ANN) was conducted to analyze the correlation between etching parameters. Levenberg-Marquart back propagation algorithm with sigmoid activation function is adopted to train the ANN model. Results show that the ANN model can predict the etching rate and etching profile accurately with 0.99154 R-Value. The relative importance of each parameter was also identified by using Random Forest algorithm. The gas flow rate of SF6 is the most significant factor of etching rate, meanwhile the ratio of SF6 to O2 flow rate plays the most important role in etching profile control.