Real-Time Optimization Design of GAAFETs Based on Machine Learning Method

This paper proposes a real-time optimization design of GAAFET based on machine learning method, using artificial neural network (ANN) to train the deep learning model, whose input and output are process structure parameters and electrical characteristic parameters, respectively. Our framework can automatically update the deep learning model with the error between the prediction and the actual, so as to realize the automatic training and decision-making. Our optimized design method has been partially confirmed on the 5nm stacked gate all around (GAA) nanosheet transistor (NST) platform, 1) the prediction time is 33962 times faster than traditional TCAD, 2) compared with the TCAD simulation results, the model achieves 99.7 % accuracy, therefore, the process time for device development is also massively shortened. At the same time, this framework is applied to the process of finding the optimal solution in the future.