We propose a supervised learning model called an adaptive weighting neural network (AWNN) that combines the advantages of statistical machine learning model and the physical TCAD model ( Aspect-ratio dependent etching (ARDE) model).
Check the link for the published work: https://ieeexplore.ieee.org/document/8558525
The outcome of the TCAD calculation is the etching depth, and the parameters used to predict the etching depth are the trench width and the etching time. Apart from TCAD, we conduct the experiment for deep reactive ion etching (DRIE) BOSCH etching process to acquire the experiment data. Pattern parameters are designed for diverse aspect ratio (AR). same as in the simulation, and therefore, the input feature values are the same for TCAD calculations and physical experiments. The etching process steps consist of pattern exposure, development, and dry plasma etching. Finally, the actual etching depth is examined with the scanning electron microscope (SEM). After experiment, the experiment data set will be fed into our artificial neural network first. Then the output result of experimental data will be combined with the TCAD data.